Dimerization Rules of Mammalian PAS Proteins

Abstract

The PAS (PER, ARNT, SIM) protein family plays a vital role in mammalian biology and human disease. This analysis arose from an interest in the signaling mechanics by the Ah receptor (AHR) and the Ah receptor nuclear translocator (ARNT). After more than fifty years by studying this and related mammalian sensor systems, describing the role of PAS domains in signal transduction is still challenging. In this perspective, we attempt to interpret recent studies of mammalian PAS protein structure and consider how this new insight might explain how these domains are employed in human signal transduction with an eye towards developing strategies to target and engineer these molecules for a new generation of therapeutics. Our approach is to integrate our understanding of PAS protein history, cell biology, and molecular biology with recent structural discoveries to help explain the mechanics of mammalian PAS protein signaling. As a learning set, we focus on sequences and crystal structures of mammalian PAS protein dimers that can be visualized using readily available software.

Background

The PAS (PER, ARNT, SIM) protein family plays a vital role in mammalian biology and human disease. The early discovery that the Ah receptor (AHR) dimer with the Ah receptor nuclear translocator (ARNT) is at the center of the adaptive metabolism of polycyclic aromatic hydrocarbons is in keeping with the idea that other PAS heterodimers mediate a variety of environmental responses in humans.^1,2 While the AHR-ARNT dimer has guided early ideas about how adaptation to oxygen and light occurs, an additional surprise has been that our understanding of AHR signaling is improving based upon advances in the structural biology and genetics of the larger mammalian PAS protein family.

In this perspective, we attempt to interpret recent studies of mammalian PAS protein structure and consider how this new insight might explain how PAS domains are employed in human signal transduction with an eye toward developing strategies to target and engineer these molecules for a new generation of ligand activatable therapeutics. Our approach is to integrate our understanding of AHR PAS protein cell biology, and molecular biology with recent structural discoveries to help provide some simple rules about the mechanics of mammalian PAS protein signaling. As a learning set, we focus on sequences and crystal structures of mammalian PAS protein heterodimers, that are either full-length proteins or that include at least two adjacent structural domains, that have known biological relevance, and that can be visualized by the non-expert in structural biology (like the authors) using readily available software suites such as PyMol (Schrodinger, NY),³ Phyre2,⁴ Alpha-Fold,⁵ Chimera,⁶ and JalView.⁷

Origins of the Term PAS

Organisms detect rapidly changing environments and regulate the adaptive physiological response for survival. This often depends on PAS sensors that employ protein–protein interactions to generate dimers and higher-order complexes that influence cellular physiology through complex networks or interactomes. Phylogenetic analyses demonstrate that PAS domains are found in all major kingdoms of life, suggesting that this structure arose from a common single-celled ancestor and quickly evolved to the needs of multicellular organisms.⁸ In many organisms, “PAS domains” are found in biological sensors that detect changes in oxygen levels, light, membrane voltage, and the presence of specific small molecules.⁹ In animals, the adaptive response to these stimuli often occurs through changes in gene expression that result from PAS dimers binding to genomic enhancer elements through the use of an associated N-terminal basic helix-loop-helix domain (bHLH)^9-11

The PAS domain has many functions. Most notably, it can serve to bind small molecules, transduce signals, and support interactions with other proteins. When these interactions are between PAS family members, we refer to this as homo-family interactions (e.g., AHR/ARNT). When these interactions are with proteins that do not harbor PAS domains, we refer to these interactions as hetero-family (e.g., AHR/Hsp90).^12-15 In this review, we further differentiate interactions as follows: PAS-PAS interactions between the same protein are “homotypic,” while interactions between two distinct PAS proteins are “heterotypic.” We also divide mammalian PAS proteins into six classes based on our current understanding of their function and primary sequence homology: alpha-class, that commonly acts as sensors of environmental cues, beta class, that act as broad spectrum dimeric partners for the alpha-class; gamma-class, that appear to function as transcriptional coactivators, delta class, that act as repressors of the alpha–beta dimers, epsilon-class, that contain enzymes, and kappa-class, that include a series of ion channel protein subunits (Figure 1).¹⁶ This chapter focuses on the homo-family interactions of mammalian alpha, beta, and delta class PAS interactions. For additional information on the underlying biology, the reader is referred to several recent reviews.^2,10,11,17

Figure 1. The PAS family of proteins in humans:

There are over 35 PAS containing proteins encoded by the human genome. In this classification scheme, we divide this family into six classes: alpha - sensors, beta - dimerization partners, gamma - coactivators, delta - repressors, epsilon - enzymes, and kappa - potassium channels. PAS proteins in color are highlighted as they are discussed in this report. The PASA and PASB domains are denoted as tan boxes. The larger PAS region as originally described is show in gray. The bHLH domain is designated as the black bar immediately N-terminal tto the PAS region. The proteins are drawn from the N-terminus (bottom) to the C-terminus (top). For the gamma class, both a PASA and PASB domain are depicted based upon a prior sequence homology analysis.¹¹ In some cases, more than one common name is provided (e.g., HIF2α, EPAS orMOP2). For the epsilon class the light green box represents unique enzyme activities and HEME designates a region known to bind heme.⁸⁶

The term “PAS domain” was initially defined through the identification of a sequence homology region between three founding proteins: Drosophila PER,¹⁸ human ARNT,¹⁹ and Drosophila SIM²⁰ (Figure 2). This initial description of an approximately 270 amino acid PAS homology region was shown to be comprised of two internal, highly degenerate repeats of about 50 amino acids each, initially designated PASA and PASB.²⁰ Due to the low level of amino acid sequence conservation in this larger PAS domain (i.e., as low as 15 % identity across family members, Figure 2), its definition has been subject to several iterations over the years, and the terminology can be confusing. For example, some define a subset of individual PAS repeats using the term “PAS/PAC domain.” The term PAC is employed to emphasize a motif of approximately 45 amino acids found near their C-terminal end of many PAS repeats.²¹ The term “PAS-fold” has also been proposed to represent this PAS-PAC region. Thus, the “PAS fold” definition describes many PAS repeats but not all.²² To add clarity here, we will use the term PAS domain to refer to the individual PASA and PASB repeats reported initially, as they can be shown to harbor considerable structural similarity (see below).

Figure 2. Classical view of the original PAS homology domain developed from the founding members of the human family, with the addition of the AHR:

Domains as first described using amino acids sequence alignments.²⁰ The PAS domain as identified, was initially defined as the entire domain harboring two repeats, A and B as well as all linkers between domains and to adjacent domains. The designation, bHLH, indicates the basic helix loop helix domain not yet formally identified in to PERs. Domains are shown to approximate scale and represent the early publication descriptions. Amino acid identity of the entire PAS domain is shown below each as compared to the AHR which was identified immediately after the founding groups and was shown to form the first mammalian homofamily dimerization partnership with ARNT.¹⁹

Definitions Based on Structure

We set out to reexamine the PAS domains in mammals by developing a learning set of sequence motifs and corresponding structures recently elucidated by X-ray crystallography and Cryo-EM (The corresponding PDB numbers and more information can be found in Table 1). Given the large number of PAS structures deposited within the PDB dataset, we attempted to employ formal criteria for inclusion in this learning set: 1) the structures must describe PAS-PAS dimerization (focusing on alpha–beta dimers and delta class repressors), 2) the proteins must come from a mammalian species, and 3) the dimer must be proven to be of biological relevance (i.e., not structural insight alone). In addition, we focused on those structures from the largest protein fragment available and did not include any isolated PAS domain structures due to the prediction that associated domains drive the character of the PAS-PAS interaction. To start, we first selected several PAS homo-family dimers from mammals: CLOCK-ARNTL, PER1-PER1, PER2-PER2, PER3-PER3 HIF2α-ARNT, HIF1α-ARNT, NPAS1-ARNT, NPAS3-ARNT, AHR-ARNT and AHRR-ARNT. During writing and review, we became aware of, and included, additional PAS dimers, e.g., HIF3α-ARNT, NPAS4-ARNT2, NPAS4-ARNT, and SGCα-SGCβ. Finally, an AHR hetero family interaction has also been elucidated (AHR-Hsp90) and is only briefly discussed since it does not describe a homofamily interaction (see below). All structures employed in this learning set are described in Table 1.

Table 1. List of homo-family structures of PAS-interactions used as a learning set.

Selected co-crystalized PAS proteins are listed in their relevant pairings (indicated by the shared PDB numbers and references). Abbreviations: AA = amino acid, h = human, m = mouse, b = cow (bos taurus), n = Nostoc punctiforme, ms = Manduca sexta. “Chain” designates the terminology used in the PDB source. NA = not applicable.

PAS Partnerships	PDB number	Class	Full Name	Chain	AA Limits	Ref.
mAHR	5V0L	α	Ah Receptor	B	37–202	68
hARNT	5V0L	β	Ah Receptor Nuclear Translocator	A	81–256	68
hAHR	5NJ8	α	Ah Receptor	A	34–209	69
mARNT	5NJ8	β	Ah Receptor Nuclear Translocator	B	85–258	69
hAHRR	5Y7Y	δ	Ah Receptor Repressor	A	38–238	75
bARNT	5Y7Y	β	Ah Receptor Nuclear Translocator	B	95–357	75
mNPASI	5SY5	α	Neuronal PAS-1	B	52–329	32
mARNT	5SY5	β	Ah Receptor Nuclear Translocator	A	99–376	32
mNPAS3	5SY7	α	Neuronal PAS-3	B	59–334	32
mARNT	5SY7	β	Ah Receptor Nuclear Translocator	A	89–364	32
mNPAS4	7XI4	α	Neuronal PAS-4	B	1–348	85
mARNT	7XI4	β	Ah receptor Nuclear Translocator	A	82–464	85
mNPAS4	7XI3	α	Neuronal PAS-4	B	3–316	85
mARNT2	7XI3	β	Ah Receptor Nuclear Translocator-2	A	61–347	85
mCLOCK	4F3L	α	Clock	A	42–360	43
mARNTL	4F3L	β	Ah Receptor Nuclear Translocator Like-1	B	71–372	43
mHIF1α	4ZPR	α	Hypoxia Inducible Factor alpha	B	15–293	56
mARNT	4ZPR	β	Ah Receptor Nuclear Translocator	A	87–321	56
mHIF2α	4ZP4	a	Hypoxia Inducible Factor 2 alpha	B	26–320	56
mARNT	4ZP4	β	Ah Receptor Nuclear Translocator	A	98–358	56
mHIF2α	4ZPK	α	Hypoxia Inducible Factor-2 alpha	B	8–328	56
mARNT	4ZPK	β	Ah Receptor Nuclear Translocator	A	87–354	56
mHIF3α	7V7L	α	Hypoxia Inducible Factor 3 alpha	B	19–317	30
mARNT	7V7L	β	Ah Receptor Nuclear Translocator	A	98–377	30
mPER1	4DJ2	δ	Period-1	A	196–471	82
mPER1	4DJ2	δ	Period-1	B	196–474	82
mPER2	3GDI	δ	Period-2	A	168–436	81
mPER2	3GDI	δ	Period-2	B	179–428	81
mPER3	4DJ3	δ	Period-3	A	119–383	82
mPER3	4DJ3	δ	Period-3	B	104–401	82
hSGCα	6JT0	ε	Soluble Guanylate Cyclase alpha	A	69–610	26
hSGCβ	6JT0	ε	Soluble Guanylate Cyclase beta	B	1–576	26
nSTHK	2P04	NA	Serine Threonine Histidine Kinase PAS domain only	A	1–107	25
nSTHK	2P04	NA	Serine Threonine Histidine Kinase PAS domain only	A	1–105	25
msSGCα	4GJ4	NA	Soluble Guanylate Cyclase PAS domain only	A	279–391	27
msSGCα	4GJ4	NA	Soluble Guanylate Cyclase PAS domain only	A	279–390	27
hAHR-HSP90-AIP	7ZUB	α	AHR bound to Hsp90 and AIP	D	1–437	28

Reexamining the PAS domain

Using visual inspection of dimer structures generated by several laboratories (our learning set), we first identified and summarized the conserved order of observed β-strands and α-helices in PAS domains from selected alpha–beta, delta-beta, and delta-delta dimers that have been structurally elucidated and for which signal transduction information is available (Figure 1, Table 1). We observe a simple pattern of α-helices and β-strands common across all of these PAS domain classes, similar to that reported by others.^22,23 As one representation of this analysis, we performed a sequence alignment of the bHLH and PAS domains of this learning set using only sequences from regions which have been structurally elucidated (i.e., any unstructured linkers or loops were not included) (Figure 3 TOP). From this analysis, we conclude that all mammalian PAS repeats begin with a β-strand and end with a β-strand (Figures 3 and 4). Despite low amino acid sequence homology, all alpha, beta, and delta class PAS proteins examined in this study conform to the following motif organization: A-β-strand, B-β-strand, C-α-helix, D-α-helix, E-α-helix, F-α-helix, G-β-strand, H-β-strand, I-β strand (denoted as Aβ, Bβ, Cα, Dα, Eα, Fα, Gβ, Hβ, Iβ)(Figure 3 and 4). This analysis leads us to a simple definition of almost all mammalian PAS repeats conforming to an earlier described concept where each PAS repeat is composed of 5 anti-parallel β-strands and four α-helices in a specific order (Aβ, Bβ, Cα, Dα, Eα, Fα, Gβ, Hβ, Iβ) (Figure 4, Right).²²

Figure 3. TOP: Sequence alignment of crystalized structures found in PAS proteins from our learning set.

Using Jalview and crystal structures from PDB, we aligned (ClustallW) the corresponding sequences (obtained as PDB/mmCIF format) to determine common structural elements observed. The major domains are the basic-helix-loop-helix (bHLH), and the PASA and PASB repeats/domains. Blue = alpha-helix in crystal; red = beta-strand in crystal. The helices and beta strands are named above according to the convention described in the text. BOTTOM: Because human SGC did not align well to other mammalian PAS proteins, its PAS domain was identified in isolation, through a comparison of PAS domains from STHK from Nostoc punctiforme and a PAS domain from the SGC from Manduca sexta. PDB numbers are provided to the right of each name.

Figure 4. Representation of mammalian PAS domains.

Using the ARNT PASB repeat (5Y7Y), the 5 anti-parallel β-strands and 4 α-helices are prominent and labelled in alphabetical order from the N-terminal side. Left is a front view of the PAS repeat which forms a cavity and contains 4 α-helices. Middle, the back view of the PAS domain showing the β-sheet. Right: Cartoon representating the PAS repeat with initial alpha helix shown in blue and the C-terminal beta strand shown in red. Unstructured regions are shown as lines, α-helices as coils and β-strands as arrows. For all depictions, terms like “Top”, “Bottom,” “Front,” and “Back,” are assigned arbitrarily and used to organize and compare structures.

The sequence and structural alignments of PAS repeats in our learning set reveals support for the conservation/importance of an alpha-helix, denoted A’α, immediately N-terminal to the Aβ strand of many mammalian PASA repeats (Figure 3). This structure is so reproducibly found near PASA, that it can be argued that it is an essential identifier/component of that structure or that the A’α helix defines PASA. Interestingly, the dimeric soluble guanylate cyclases (SGC) which harbor a single PAS repeat,²⁴ and included here as epsilon members of the PAS family, provide an early demonstration of the importance of A’α. While we have tried to focus on mammalian PAS protein structure, the importance of an A’α helix to the dimerization of SGCs was first structurally revealed in a bacterial serine-threonine-histidine-kinase (STHK) from Nostoc punctiforme (2P04),²⁵ and recently reproduced in the human α- and β-subunits of SGC (6JT0).²⁶ To demonstrate this idea, we aligned these sequences and structures with an orthologous PAS repeat found in the tobacco hornworm (Manduca sexta)(4GJ4).²⁷ This presentation allows clear definition of the PAS domain boundaries and sequences within the human SGC proteins (Figure 3 Bottom). This analysis also documents the role of the A’α helix in an epsilon family member, and supports its importance in SGC subunit dimerization. Finally, the alignment reveals that the structural organization of these two epsilon members, and perhaps domains from other classes of PAS proteins, may not perfectly conform to the description that so robustly applies to the alpha, beta, and delta PAS repeats. In this regard, while the A’α, Aβ, Bβ, Cα, Dα, Eα, Fα, Gβ, Hβ, Iβ, have highly orthologous structures within the PAS domains of SGCs, their sequences and structures can be distinctive. For example, an additional beta strand can be observed between the E and F helices of SGCs (denoted in Figure 3, BOT-TOM as βX).

The structure of PAS domains provides at least two distinct functional properties to a protein. The organization of the α-helices, β-strands and intervening loops, create multiple dimerization surfaces for unique homo and hetero dimeric interactions, and support cavities that bind and transduce signals from small ligands like those observed in the PASB domains of the AHR,^28,29 HIF3α,³⁰ HIF2α,³¹ and perhaps a variety of other family members.³² The presence of more than one PAS domain in a given protein may provide selective advantages allowing independent diversification of each PAS repeat’s function. This may allow increased dimerization specificity, increased dimerization avidity, capacity to bind a unique spectra of small molecules, and increased capacity to transduce cellular signals. The presence of two PAS domains within a single protein thus provides potential for independent evolution of each repeat. While not discussed here, a number of excellent reviews describe the roles of PAS mediated dimerization and PAS as a small molecule binding pocket in a broad spectrum of nonmammalian proteins where these domains were first studied (even though the term PAS arose later from their rediscovery in animal models).^8,9,33,34

Patterns of Dimerization

To simplify the discussion of the PAS domain structure, we arbitrarily define the β-sheet as the “back” of the PAS domain structure and the cavity as the “face” of the structure (Figure 4).³⁵ Similarly, we define the loop between the H and I strands (“HI Loop”) as the top and the loop between the G and H strands (“GH Loop”) as the bottom. This structural/visual-driven definition of PAS repeats complements the description of this region via sequence alignment. Although PAS secondary and tertiary architecture are conserved, primary sequences can be difficult to recognize by homology searches at the amino acid level. For example, between mouse and human, PASA motifs share about 35 % homology, while PASB motifs show about 31 % homology.² Interestingly, there is less than 20 % homology between PASA and B.

With a stated definition of the PAS domain, we turned our attention to the ways in which these domains interact, focusing on homo-family interactions (i.e., PAS-PAS) between alpha and beta family members. To accomplish this, we analyzed each dimeric pair from our learning set visually and informatically using PyMol, UCSF-Chimera, Phyre2, and the CLUSTALL algorithm embedded in JalView.

We had two objectives: First, to identify common modalities of PAS protein interactions in mammals. Second, to challenge the idea commonly used in the PAS field of research (including by our own laboratory) that depicts their signal transduction pathways as a symmetrical pairing of bHLH-PASA and PASB domains associated with the active state of the dimer (Figure 5). This common representation of PAS signal transduction has its roots in the symmetry of known bHLH structures and other homodimeric nuclear receptors such as the glucocorticoid and estrogen receptors that were co-emergent with the origins of PAS discovery.^36-39 In addition to their similarities as ligand-activated transcription factors that act through dimerization to activate gene expression, the molecular and structural biology of the nuclear receptor family also presaged the importance of common dimerization partners in the PAS family. Just like the Retinoid-X receptor acts as a partner to an array of nuclear receptors, the beta-class PAS members (e.g., ARNT) form productive dimers with a variety of alpha-class members,^1,40 There-fore, we systematically analyzed our learning set for the mechanisms by which PAS domains interact, searching for similarities and symmetry.

Figure 5. Classical schematic for alpha–beta dimerization.

As sensors, the class alpha member senses an environmental signal and is stabilized or transformed, and moves to the nucleus through the activity of a revealed nuclear localization sequence. In the nucleus, the alpha–beta members partner. The resultant dimeric transcription factor, positions the basic helix of the two bHLH domains to recognize congnate enhancers of target genes. This classical drawing represents the idea that similar surfaces of PASA and PASB domains interact and that they all interact in the same fashion. Red stars are used to denote assumed points of contact.

The CLOCK:ARNTL Dimer (4F3L):

One of the earliest dimers to be structurally resolved from the mammalian PAS family, contains the bHLH, PASA and PASB domains (but not C-terminal variable region, or the bound DNA enhancer), from the CLOCK-ARNTL dimer (ARNTL is also known as BMAL1⁴¹ or MOP3^1,42).⁴³ This dimer is a transcriptional activator found in a variety of cell types, including the suprachiasmatic nucleus of the hypothalamus.^2,44 It regulates daily physiological and behavioral activities for the 24-hour circadian cycle. In a day, circadian inputs such as light, food, or hormones influence levels of ARNTL and CLOCK, increasing the amount of dimer formed, and its subsequent binding to cognate genomic enhancers for various output genes, including repressors of the pathway, such as the delta class PAS proteins PER1–3. The CLOCK, ARNTL, and PER mRNAs and proteins oscillate throughout the active and inactive phases of the day. The levels of CLOCK-ARNTL are highest during the active phase, while PER levels increase in the inactive/sleeping phase. This increase in a PER-CRY complex facilitates their translocation to the nucleus and subsequent disruption of the ARNTL-CLOCK-ARNTL complex. Later in the cycle, CLOCK-ARNTL is again formed and the process re-starts. Almost all mammalian cells express clock genes that create oscillations in a similar manner. In some cells, paralogues of CLOCK and ARNTL may also operate (CLOCK paralogs known as NPAS2⁴⁵ or MOP4¹ as well as ARNTL paralogs known as either ARNTL2, MOP9⁴⁶ or BMAL2⁴⁷). In sum, these CLOCK-ARNTL dimers sense environmental inputs such as food or light and serve as a central transcriptional engine that synchronizes circadian rhythms in essentially all cells.

The CLOCK-ARNTL crystal structure (4F3L) provided one of the first looks at a mammalian PAS protein dimer, revealing a complex with multiple aspects of symmetry (Figure 6).⁴³ In accordance with our classification scheme, CLOCK is an alpha-class member, while ARNTL is beta-class (Figure 1).¹¹ Our own initial models assumed the three highly conserved domains – bHLH, PASA, and PASB – would interact predictably and consistently (i.e., as in Figure 5). That is, the CLOCK bHLH interacts with the ARNTL bHLH, the PASA with the PASA, and the PASB with the PASB. More-over, PASA-PASA interactions would be like PASB-PASB interactions. Evidence from the CLOCK-ARNTL structure demonstrated that our early simplistic model is incorrect. In 3D-space, bHLH-bHLH and PASA-PASA interactions are medial while PASB-PASB interactions are lateral. Additionally, the PASB-PASB dimer is stacked on top of PASA-PASA. In both CLOCK and ARNTL, PASA-PASB contacts are also made so that PASAs interact with both PASBs (Figure 6). We refer to this region of the PASA dimer as the “PASA Platform” as it provides a surface for PASB dimers to interact. A final note of distinction between the PASA repeats is the observation that the CLOCK bHLH and its PASA domain interact directly, while the corresponding regions in ARNTL do not.

Figure 6. PAS Repeat interactions from the CLOCK-ARNTL crystal structure (4F3L):

ARNTL is shown in dark blue. CLOCK is shown in orange. Domains are noted in figure.

Notably, the PASA domains of CLOCK and ARNTL barely interact. In fact, the dimerization of these domains is primarily supported by the accessory α-helix of PASA domains, A’α, that lies between the bHLH and the PASA repeat.⁴³ While this contribution to dimerization by A’α was predictable based upon earlier work with SGC (see above), the CLOCK-ARNTL structure was the first mammalian PAS pair to document the importance of this PASA accessory domain in mammalian alpha–beta class dimerization. Dividing CLOCK and ARNTL with an imaginary midline allows one to see that the A’α-helix of one protein crosses over to establish extensive contacts with the “back” of the PAS-A β-sheet of the other, while also interfacing with one another (Figure 6). Outside of the A’α-helix, the “backs” of the PASA β-sheets point toward this imaginary midline and the “faces,” containing the 4 helices, point outward. Therefore, the A’α-helices on both proteins play a key role in stabilizing PASA-PASA interactions in mammalian systems.

The HIF2α-ARNT and HIF1α-ARNT Dimers (4ZPK and 4ZPR):

Oxygen homeostasis is regulated by the alpha-class PAS proteins known as HIF1α, HlF2α, and HIF3α and their beta-class partners ARNT and ARNT2² (and probably ARNTL and ARNTL2⁴⁸). The HIFαs are sensors that are upregulated through protein stabilization under hypoxic conditions and heterodimerize with their beta-class partners, like ARNT, to transcribe target genes responsible for survival and adaptation to low oxygen levels.⁴⁹ Under normoxic conditions, HIFαs are constantly produced and degraded, however, during hypoxic conditions, enzymes that hydroxylate HIFs lose activity in relationship to their oxygen requirement. This hydroxylation in the variable region permits recognition and ubiquitination by Vh1, an E3 ubiquitin ligase.⁵⁰ Without ubiquitination, HIFα concentrations increase in the cell, translocate to the nucleus, bind ARNT, and increase transcription of target genes with roles in adaptation to low oxygen. The HIF1α and HIF2α proteins are the most studied members of this group and most structurally similar, while the HIF3α protein is a more enigmatic member that may act as a pathway repressor or modifier.⁵¹

The ARNT protein is the prototype beta-class PAS protein because it has been shown to dimerize with at least 12 of the over 30 known mammalian PAS proteins. This protein class is largely resistant to upregulation (e.g., hypoxia regulation), and can be thought of as a stably expressed protein that awaits its activated partners, as in cases of the HIFαs and the AHR. Interestingly, in the nucleus ARNT has been shown to homodimerize, although the biological relevance of this interaction is uncertain.^52,53 Out-side of PAS proteins, the ARNT protein is also known to interact through hetero-family modes with various coactivators, like TACC3,⁵⁴ and steroid receptors like the estrogen receptor alpha (ERα).⁵⁵

The structures of the HIF2α-ARNT (4ZP4 and 4ZPK) and HIF1α-ARNT (4ZPR) dimers add support for the commonality of the dimerization modes that were revealed from the CLOCK-ARNTL structure.⁵⁶ One repeated theme is that both HIF1α-ARNT and HIF2α-ARNT dimers also use A’α helices to stabilize PASA-PASA interaction (Figure 7, Left).⁵⁶ Like in CLOCK-ARNTL, the A’α helices cross the midline to interact with each other as well as the back of opposing beta sheets. A second theme we also see repeated from CLOCK-ARNTL is that PASB-PASB in the HIFα-ARNT interactions employs a different mode of dimerization than PASA-PASA interactions. Again, unlike PASA, PASB-PASB interactions are front to back with direct interactions between residues within the respective PASB repeats (Figure 7, Middle). Finally, we also see the additional feature where the PASA-PASA dimerization creates a platform for the PASB dimer to reside. This platform interaction is subtly different from the CLOCK-ARNTL interaction. In the HIF1α-ARNT and HIF2α-ARNT dimer, it is primarily the PASB repeat of HIF1α residing on the PASA platform with ARNT being linearly extruded with less interaction to any PASA repeats (Figure 7 Right, and see more below).

Figure 7. PAS Repeat interactions from the HIF2α-ARNT and the HIF1α-ARNT crystal structures (4ZP4 and 4ZPR, repectively):

Top: dimeric structure of the ARNT-HIF2α structure and its subdomains. HIF2α is shown in yellow and ARNT is shown in gray. Bottom: dimeric structure of the ARNT-HIF1α structure and its subdomains. HIF1α is shown in green and ARNT is shown in gray.

Some new ideas also arise from a comparison of the three structures described so far (i.e., 4F3L, 4ZPK and 4ZPR). First, are some distinct PAS-PAS configurations. In the HIF1α:ARNT dimer, the Fα on HIF1α plays an important role for PASB-PASB stabilization, as well as the apparently disordered region between residues 247–279.⁵⁶ On ARNT, the Hβ, Iβ, and the Hβ-Iβ loop establish the necessary connections with HIF1α. Since HIF1α and HIF2α are highly homologous at the PAS repeats and throughout the domain, it is expected that the protein interfaces are the same for both proteins. Interestingly, for HIF2α the Aβ, Bβ, Aβ-Bβ loop, Fα helix, and the disordered region and located outside of PASB, all interact with ARNT PASB.⁵⁶ This is not the case with HIF1α, therefore sequence homology does not appear to be the only dictate to predict modes for PAS-PAS interactions. Second, in the HIF2α:ARNT dimer, additional contacts are observed between PASA and DNA; specifically N184 and K186 on the Gβ-Hβ loop of the HIF2α (data not shown).⁵⁶ Currently, this interaction is unique among crystallized PAS dimers. Third, the ARNT PASA repeat not only interacts with the HIF1α and HIF2α PASA, but also interfaces with the HIF1α and HIF2α PASB repeats (Figure 7, Right and see below). To establish these connections, the ARNT PASA relies on what appears to be the Fα helix, Gβ, and Hβ strands while HIF1α/HIF2α use a portion of the linker between its PASA-PASB repeats and the Hβ and Iβ strands.⁵⁶

The NPAS1:ARNT and NPAS3:ARNT Dimers (5SY5 and 5SY7):

The neuronal NPAS1⁴⁵ (also known as MOP5¹) and NPAS3⁵⁷ proteins are highly expressed in the nervous system and mutations have been linked to schizophrenia and autism.^45,58-59 While fewer details are known about the details of their respective biological pathways or their functional domain maps (as compared to HIF, CLOCK or AHR) they represent an exciting avenue of future research with clear implications for human disease and therapy. The NPAS members appear to fall into the alpha-class of the PAS family^2,16 and their study is an excellent representation of where a partner prediction strategy based on sequence homology and class assignment is valuable (i.e., all class-alpha dimerize with class-beta etc., Figure 1).^2,48,52 The NPAS1, NPAS2, NPAS3 and NPAS4 all appear to be alpha-class members and potentially relevant partners of the beta-class members ARNT, ARNT2, ARNTL, and ARNTL2. Importantly, the mechanisms by which these classes of proteins might distinguish binding partners is yet uncertain, and currently is largely perceived to be related primarily to cell specific coexpression (i.e., which beta-class partner is expressed in the cell type where the alpha-class protein is relevant, see more below).

Like the other alpha–beta class dimers, the PASA-PASA connections of both NPAS1-ARNT (5SY5) and NPAS3-ARNT (5SY7) are supported by adjacent bHLH interactions and then stabilized by their A’α helices and beta sheets (Figure 8).³² Also similar, the ARNT PASA interacts with both the PASA and PASB of NPAS1 and 3, and the PASB of ARNT only interacts with the PASB of NPAS1 and 3 (i.e., again the PASB-PASB dimer sits on a PASA-platform and the PASB of ARNT extrudes outward, Figure 8, and more detail in Figure 9). Examination of the PASA-platforms from these and earlier structures reveals some subtle differences. For example, in the NPAS1-ARNT dimer, the NPAS1 PASB interaction with ARNT PASA relies on the linker between ARNT PASA-PASB, the Fα helix, Gβ, Hβ, and Hβ strands; while the NPAS1 PASB uses the linker between PASA-PASB, Gβ, Hβ, Hβ, and a loop near Gβ-Hβ (AA: 366–369)(Figure 9).³² Although the PASA and PASB repeats of NpAS1 appear to be in close contact, only the Hβ on PASA and Gβ on PASB appear to interconnect. For the PASB-PASB configuration, the NPAS1 uses helices Eα and Fα and the Aβ-Bβ loop as well as AA 406–421, which appear to be a loop followed by a helix outside the PASB domain.³² The ARNT counterpart requires more regions to coordinate interaction: Bβ, Bβ-Cα loop, Cα, Dα, Hβ, Hβ-Hβ loop and the Hβ.

Figure 8. PAS Repeat interactions from the NPAS1-ARNT and the NPAS3-ARNT crystal structures (5SY5 and 5SY7, respectively):

Top: dimeric structure of the ARNT-NPAS1 structure and its subdomains. NPAS1 is shown in silver-red and ARNT is shown in gray. Bottom: dimeric structure of the ARNT-NPAS3 structure and its subdomains. NPAS1 is shown in pink and ARNT is shown in gray.

Figure 9. Interactions in the PAS Platform:

Top Left, The ARNT PASA interacting with both repeats from NPAS1 (5SY5). Top Right: Most of the interactions apparently require a portion of the disordered region as well as the β-sheet. Some connections alse exist between NPAS1 PASA and B repeats (5SY5). Bottom Left: The ARNT PASA interacting with both repeats on NPAS3 (5SY7). Bottom Right: Most of the interactions apparently require a a portion of the disordered region as well as the beta sheet. Some connections exist between NPAS3 PASA and B domains (5SY7).

In NPAS3-ARNT, the ARNT PASA employs the Eα, Fα helices, Gβ, Gβ-Hβ loop and Hβ for PASA to establish contact with the NPAS3 PASB. The NPAS3 PASB also utilizes flexible regions (linker and the loop following PASB) to establish contact with the ARNT PASA, as well as some beta strands (Hβ and Hβ).³² Interestingly, the NPAS3 PASA has very little contact with its PASB, however both repeats interact with the ARNT linker. For PASB-PASB interconnectivity, NPAS3 relies on the Aβ-Bβ loop, Eα, Eα-Fα loop, Fα and a helix out-side of PAS B; ARNT uses all structures except Eα, Fα and Gβ and the loops between them (Figure 9). This is similar to what is observed in the NPAS1-ARNT interactions and is supported by the high level of sequence homology between NPAS1 and NPAS3.

The AHR:ARNT and AHRR:ARNT Dimers (5V0L and 5Y7Y):

The prototype mammalian PAS heterodimer, is the AHR and its partner ARNT.^60,61 This is the first mammalian bHLH-PAS protein heterodimer known to be activated by ligand binding within PASB.^19,62 Signaling by this system begins with the unliganded receptor held in a cytosolic complex with a dimer of Hsp90, a single molecule of P23, and a single molecule of A!P (also known as ARA9).^13,15,63,64 When ligands diffuse through the cell membrane into the cytosol, they encounter the AHR. Once activated by ligand, the AHR changes associations with its cellular chaperones and a nuclear localization sequence (NLS) is revealed leading to the complex’s increased concentration in the nuclear compartment where it dimerizes with the constitutively nuclear ARNT.^19,65 The AHR-ARNT heterodimer employs its bHLH domains to bind enhancer elements regulating genes encoding various cytochromes P-450 dependent monooxygenases (i.e., CYP1A1, CYP1A2 and CYP1B1), as well as the “aryl hydrocarbon receptor repressor” (AHRR).^66,67 These four genes are believed to play important roles in downregulating the AHR response, with the CYPs displaying roles in the metabolic clearance of xenobiotic ligands, while the AHRR acts as a repressor of AHR-ARNT mediated transcription.

The bHLH domain and PASA repeat of the AHR-ARNT dimer bound to DNA has been crystallized and its structure solved by two laboratories (5V0L and 5NJ8).^68,69 The AHR PASB repeat has proven difficult to crystallize included within an AHR-ARNT dimer. This is presumed to be the consequence of the AHR’s PASB requiring numerous chaperone associations to prevent aggregation.⁷⁰ Hence, several in silico predictions of the AHR-ARNT dimer have been reported.^29,71,72 Like other PAS dimers described above, PASA-PASA interactions are facilitated through the A’α helix (Figure 10). At the present time, we do not know how the AHR or the ARNT PASB repeats interact, but future studies of this interaction may be guided by the assembly of the other alpha–beta pairings or future solutions based on CryoEM.²⁸

Figure 10. PAS Repeat interactions from AHR-ARNT crystal structures.

Left (5V0L):: AHR PASA and ARNT PASA dimer isolated from complete structure. Aquamarine blue is AHR, dark gray is ARNT. Middle (5V0L): AHR-ARNT PASA interactions. The full-length AHR proteins has not been crystalized, however the resolution of its bHLH and PASB in dimeric form with ARNT has been carried out. Right (7ZUB): The PASB of the AHR: (aquamarine blue) with indirubin ligand bound internally (green). Structure includes the dimer of Hsp90 (red) with magnesium ions (green), as well as the AIP monomer (chocolate brown).

For objectivity and to avoid the influence of crystal structures that do not reflect the composite influence of multiple dimerization domains, these attempts to understand AHR-ARNT signaling employed structures with more than one dimerization domain from a pairing with known biological import (i.e., bHLH and PASA, but not PASB, for an AHR-ARNT dimer with DNA). They also reflect a limitation of our “learning set approach,” as the earliest structure of the AHR PASA domain was not included because it did not meet our criteria of at least two dimerization domains included in the structure or strongly supported biological relevance. Nevertheless, an earlier structure of the isolated PASA domain of the AHR demonstrated the importance of the A’α helix on AHR-ARNT dimerization. Moreover, a consideration of these structures with the PASB domains of CLOCK-ARNTL and HIF2α-ARNT supports an the important idea, that dimerization modes of the PASA and PASB domains between AHR and ARNT will be distinct.⁷³

Interestingly, the PASB from AHR, with and without ligand (7ZUB and 8H77, respectively), has recently been solved through Cryo-EM technology providing an important early example of mammalian PAS heterofamily interactions.^28,74 These structures are interesting for three reasons. First, they demonstrate that the back of the AHR PASB domain (i.e., the β-sheet) docks within the Hsp90 dimer and is flanked by an association with AIP. These associations may prevent aggregation and inhibit ligand independent ARNT dimerization by blocking AHR’s PASB from binding to the PASA-platform that seems essential for all alpha–beta dimers shown above. Second, the AHR’s front is partially solvent exposed. This orientation provides ligand access to the internal fold and potential for allosteric responses once ligand enters. Ligand-induced folding changes may also reveal the NLS through movement of Hsp90 or AIP associations.⁶⁵ Finally, revealing of the AHR’s face through ligand induced changes in chaperone interactions may expose the PASB repeat in a manner available for ARNT PASB dimerization.

The AHRR:ARNT (5Y7Y) dimer provides one example of the associations of an alpha-delta class dimer (Figure 1).⁷⁵ The AHRR bears considerable sequence homology to the AHR, binds ARNT with high affinity, and through positioning of bHLH domains, its ARNT dimer binds the same enhancer sequence as the AHR-ARNT dimer.⁶⁷ Importantly, the AHRR lacks a PASB domain and is unable to bind ligands. Although its sequence homology is closest to the full-length AHR, its assignment to the delta-class arises from two origins. The first is that the AHRR, like the PERs, represents a PAS protein that is missing a domain found in most other family members (e.g., for the AHRR it is the PASB domain, while for the PERs it is the bHLH). Another functional similarity to PER is that transcription of the AHRR gene is driven by enhancer elements for the activated AHR-ARNT. This is similar to the circadian pathway where PER is upregulated by the corresponding alpha–beta dimer (CLOCK-ARNTL) to provide feedback inhibition. Unlike PER, the mechanisms for AHRR mediated repression are postulated to reside in two properties⁶⁷: First, the capacity of the AHRR to dimerize directly with ARNT, thus reducing the amount of available ARNT for dimerization and reducing DNA binding of the productive complex. Second, the C-terminal half of the AHRR harbors a functional repressor domain, as opposed to a transcriptional activation domain, common to most bHLH-PAS proteins.⁶⁷ Like the AHR-ARNT dimer, the AHRR-ARNT dimer stabilizes its PASA-PASA interactions through their A’α helices similar to that described above (Figure 11).

Figure 11. PAS Repeat Interactions between the AHRR and ARNT (5Y7Y):

ARNT is shown in gray. AHRR is in dark green.

There are numerous differences between AHR and AHRR PASA domains that may give rise to subtle differences in their binding modes, as well as the repressive activity of the complex. First, unlike the AHRR, the Eα helix of AHR does not appear to have direct contact with the ARNT. Second, an extra helix follows the Fα helix on the AHRR, referred to here as F2α, and this provides a unique contribution to ARNT PASB interactions along with the Fα helix and the Eα-Fα loop. Third, the Hβ-Hβ loop is 17 amino acids, and, if similar to HIF2α, one can speculate that it may contact DNA, perhaps guiding the dimer to unique enhancers. Fourth, the ARNT PASB no longer has a parallel PASB domain in its alpha-partner. Thus, it establishes contacts with the AHRR PASA through its linker, Aβ, Bβ, Bβ-Cα loop, Fα-Gβ loop, Hβ, Hβ-Hβ loop, and Iβ. All the deviations observed from comparing the AHRR-ARNT and AHR-ARNT dimers are of interest because they may provide clues about how AHRR may conduct its repressor function. Notably, such insight may ultimately require a structural description of the variable C-termini of PAS family members. This task has not yet been accomplished for any member of this family to date.

The PER1, PER2, and PER3 Homodimers (3GDI, 4DJ2 and 4DJ3):

As noted for AHRR and perhaps HIF3α, feedback repression plays a significant role in gene regulation by mammalian PAS proteins. The PERs are the first example of this importance, with seminal work coming from the Drosophila model.⁷⁶ Early ideas were that the repressive activity of PERs on the molecular clock occurs because this PAS protein, without a DNA binding domain, replaces one of the DNA binding partners in the CLOCK-ARNTL dimer to yield an unproductive complex.^48,77 Evidence now points to mechanisms of action where PER-mediated repression influences circadian cycling in mammals through its heterofamily interactions with several proteins, most notably the CRY1 and CRY2.^44,78 As the result of CLOCK-ARNTL binding to the enhancers for these genes, transcription of the PERs and CRYs is upregulated, and the proteins are highly expressed early in the active phase. Upon interaction in the cytosol, the heterofamily dimer then translocates to the nucleus where CRY binds to the CLOCK-ARNTL dimer, and PER removes CLOCK-ARNTL from its DNA enhancers. Importantly, structures are not yet available to show how the PAS domain influences these interactions, but this represents an important objective for future research.

In humans, three known PER paralogs exist (PER1, PER2, and PER3). The PERs are interesting because they homo- and heterodimerize as part of their nuclear translocation and repressor activity.^79,80 This ability to homo- and heterodimerize makes PERs one of the more unique interactions in the PAS family. For example, Drosophila PERs have two α-helices C-terminal to PASB and similarly located α-helical domain was found in mouse PERs, were found to be important for PAS domain interactions.^81-83 To name accessory motifs in accordance with the terminology used previously, we will refer to this region as Jα as a counterpart to our use of A’α helix to describe the accessory domain immediately N-terminal to the PASA domain. Interestingly, the Jα region on PERs appears to contain a nuclear export signal (NES).^81,82 Additionally, where PER1 and PER3 have the A’α helix, PER2 does not; hence multiple unique dimerization modes are possible among PER-PER interactions. We discuss only a select set of these interactions to show examples of unique dimerization modes that have not been represented above.

Multiple dimers/oligomers of PERS have been elucidated using X-ray crystallography. One dimerization mode is represented by PER2 homodimers (3GDI).⁸¹ In PER2 homodimers, there is minimal interaction between both PASA domains. This may be related to the lack of an A’α helix in PASA. Without PASA-PASA interactions driven by this accessory helix, PER2 homodimers rely more on intermolecular PASA-PASB and PASB-PASB interactions (Figure 12, Right).⁸¹ Direct contacts between PASA on molecule one and PASB on molecule two are managed through the loop akin to Dα-Eα in PASA and the Jα helix on the C-terminal side of PASB. Interestingly, this interaction facilitates direct contact between the PASA repeat and a nuclear export signal (NES) located on the first Jα helix.⁸¹ This phenomenon is not unusual as NESs are typically found in amphipathic α-helices, where non-polar residues are buried within the dimerization interface, and the last residue tends to be solvent exposed for protein export.⁸⁴ For PASB-PASB interfacing, the Hβ, the Iβ strands, and the Hβ-Iβ loops are employed (Figure 12, Left, Middle). Aromatic amino acids on the Hβ and the Iβ strands of one molecule establish stacking interactions with the same amino acids on the second molecule. These residues are important for hydrophobic stacking interactions and are usually buried in the protein core. The Hβ-Iβ loop on molecule one harbors a TRP that establishes contact with the Iβ strand at a GLY on molecule two.⁸¹ Additionally, TRP419 also interfaces with a TRP412 on the Iβ strand.⁸¹

Figure 12. PAS Repeat Interactions within the PER2 Homodimer (3GDI):

Left: Mammalian PER2 homodimers use Hβ, Iβ and and the HI loops to stabilize PASB repeat interactions. Middle: Stacking interactions through PHE (blue) and TRP (red) residues on Hβ and Iβ strands and HI loop are shown. The crystal also shows the position of the Ja that harbors a nuclear export signal (shown for molecule-2 Left and Middle). Right: complete crystal structure from which subdomains were derived with repeats identified by circling. Molecule 1 and 2 are shown in different shades of gold.

The second dimerization mode within PERs is observed in PER1 (4DJ2) and PER3 homodimers (4DJ3),⁸² whereby PASB-PASB contacts are made, as well as PASA-PASA contacts (Figure 13). Although PER1 and PER3 proteins show A’α helices, in these crystal structures, they are not employed for PASA-PASA contacts as in many alpha–beta dimers.⁸² Instead, PER1 and PER3 homodimers rely on Fα helices (referred to in other reports as αC).⁸² Intriguingly, the helices don’t cross over to interact with the back of the beta sheets as in other PASA dimers; instead, the helices line up and establish contact among themselves to stabilize dimerization (Figure 13). For PER1 PASB-PASB interactions, like in PER2 homodimers, the backs of each β-sheet align toward the midline and interact. The helices and a few β-strands are observed on the outer perimeter. Much like PER2, the Hβ, the Iβ strands, and the Hβ-Iβ loops play prominent roles in homodimerization. In PER1 homodimers, the TRP448, on the Hβ-Iβ loop of molecule-1, interfaces with GLY455 on the Iβ strand on molecule-2. Additionally, TRP448 on molecule-1 interacts with TRP441 on molecule 2. These positions are analogous in PER3, whereby TRP359 on the Hβ-Iβ loop of molecule-1 interacts with GLY368 on the Iβ strand and TRP352 on the Hβ strand of molecule-2 (Figure 13).

Figure 13.

PAS repeat interactions in PER1 and PER3 homodimers (4DJ2 and 4DJ3, respectively).

Similarities and differences

This visual analysis of our mammalian learning set supports the idea that structural similarity, and similar dimerization modes, are at play across the mammalian PAS protein family (Figure 14). This is especially true when considering alpha–beta-class dimeric pairs. In support of an argument for similarity, we present side-by-side representations of the five structural solutions of alpha–beta dimers that include both PASA and PASB domains and the bHLH bound to cognate DNA enhancers (Figure 15). What is observed is a striking similarity of overall topology and structural organization across all pairings.

Figure 14. Modules of mammalian PAS proteins:

Cartoon represents the domains described in this review. The bHLH is N-terminal and found in many PAS proteins. The PAS repeat with initial a-helix shown in blue and the C-terminal β-strand shown in red. The A’α and Jα-helical domains are shown as coils. Loops and unstructured regions are shown as lines, α-helices as coils and β-strands as arrows. Linkers and unstructued regions are comonly found between each domain. The C-terminal hypervariable region is depicted as a random sequence and represents the domain where the greatest informtion gain can be attained through future structural experiments.

Figure 15. Similarities of known alpha–beta PAS protein interactions.

The five alpha–beta dimers with structural solutions that included the bHLH, PASA, PASB and the cognate DNA element were chosen for comparison (Chimera Multialign tool). The names of each dimer and their PDB numbers are given above. The beta class protein is shown in cobalt and the alpha class is show in tan. Each was oriented so that the beta protein’s basic-alpha helix (of the bHLH) was pointed toward the viewer and positioned to be looking straight down the DNA helix.

While at a gross level, the depiction in Figure 15 is supportive of the idea that all alpha–beta class dimeric pairs employ their bHLH domains, their A’α helices, PASA domains, PASA platforms, and PAS B domains similarly, there are also some important subtle distinctions that are consequential. One example comes from a comparison of the NPAS4-ARNT (7XI4) and NPAS4-ARNT2 (7XI3) complexes.⁸⁵ Despite these two structural solutions arising from the same lab, and both including the same cognate DNA element, significant differences were observed between the PASA domains of these two dimeric pairs, especially across the beta-class members (Figure 16, Left). This distinction led to the proposal that the different utilization of PASA surfaces may influence signaling through alteration in avidity for DNA binding or presentation of unique interaction surfaces for distinct transcriptional coactivators. Such mechanisms could explain how one alpha-class PAS protein, NPAS4 could employ different beta-class partners to influence multiple gene batteries and explain why distinct pairings are important in the same cell type.

Figure 16. Structural differences between PASA domains within our learning set.

To emphasize differences between PASA domains, the software program Chimera was employed along with its integrated alignment tool. Left: Alignment of the PASA dimers of ARNT2-NPAS4 and ARNT-NPAS4 (7XI3 and 7XI4, respectively). Right: Alignment of the PASA dimers of ARNTL-CLOCK and ARNT-HIF3α (4F3L and 7V7L, respectively). Beta class PAS proteins are shown in blues and alpha class are shown in tans.

Our own early pairing experiments, when most mammalian PAS proteins were orphans, indicated that not all alpha–beta pairs dimerize to form functional complexes.^1,46,48 In this regard, our early two-hybrid, coprecipitation, and transient transfection studies indicated that the beta-class ARNT was a partner for the AHR and HIFs but not CLOCK. Conversely, the beta-class ARNTL (MOP3 in our early studies) was a functional partner with CLOCK and HIFS but not the AHR. These observations indicate that while the overall topology of PAS pairings may be grossly similar, fundamentally important differences in their interaction surfaces are imparting distinct pairing specificities. As one example of this idea, a comparison of the PASA pairings of CLOCK-ARNTL and HIF3α-ARNT (4F3L and 7V7L, respectively) are expanded to demonstrate the distinctive differences in their conformations and the specific residue availabilities of the PASA surfaces of these two alpha-class partners (Figure 16, Right).

Synthesis and Conclusions

This perspective began with the simple question; “how do we define a human PAS domain?” It also has its roots in the old idea, that the functional roles of PAS dimers can be predicted based upon their domain structure. These underpinnings led us to attempt to integrate our knowledge of the biology of mammalian PAS protein signaling with a visual inspection of the emerging mammalian PAS dimer structures provided by many talented laboratories over the last decade. We hope these summaries of structural data and our own views will generate new testable ideas about the mechanisms of several important mammalian signal transduction pathways. Moreover, we hope these summaries can provide direction for related therapeutic approaches and allow the development of tools for detecting related environmental or endobiotic stimuli.

To conclude, we summarize what we have learned about PAS protein biology and what we see as important questions:

1. Definitions of PAS repeat: There are many ways to define a PAS domain. The one we find supported by this visual inspection of structural solutions and sequence alignment is that all mammalian PAS repeats begin at its Aβ strand and ends with its Iβ beta strand. Furthermore, almost all mammalian PAS domains have the fol-lowing structure: Aβ, Bβ, Cα, Dα, Eα, Fα, Gβ, Hβ, Iβ.

2. Alpha-Beta dimers: Alpha-beta dimers follow some simple interaction rules. These include i) N terminal bHLH domains stabilize PAS dimers and position basic alpha helices within the major groove of target DNA, ii) PASA-PASA domain interactions are stabilized in a back-to-back configuration by an A’α helix accessory domain, iii) A PAS-platform established by the PASA-PASA interaction, holds the PASB repeat of the alpha-class partner and extends the beta-class PASB domain via a “front to back” and highly parallel PAS-PAS interaction.

3. Delta class protein dimers: The PER proteins and the AHRR break many of the rules developed from the study of alpha–beta dimers. While AHRR-ARNT dimers incorporate the A’α helix to support PASA-PASA dimerization, their PAS plat-forms are unique and unlike those observed in any other PAS dimer to date. Similarly, while some PER proteins harbor an A’α helix adjacent to their PASA, this helix is not employed in PASA-PASA interactions. Here, aromatic residues proximal to the Hβ-Iβ loop provide the most significant support for dimerization. Similarly, while PER1 and PER3 have A’α helices they have not yet been shown to be employed in PAS A interconnectivity; instead, they rely on their Fα helices for stabilization. These data lead us to speculate that the PER dimerization structures reported to date may only be a subset of those related to the biological activity of these proteins.

4. Importance of accessory domains: Accessory domains are important in PAS dimerization: Domains such as the bHLH, A’α and Jα are critical, if not essential, in many PAS-PAS interactions. In fact, PAS domains may be defined by their proximity to accessory domains.

5. Length of PAS repeats: Across the mammalian PAS family, the PASA repeat is longer in amino acid sequence than the PASB repeats. This is due to longer Fα-Gβ, Gβ-Hβ, and Hβ-Iβ loops.

6. Unstructured and unsolved regions of PAS repeats are important: The observation that unstructured and unsolved regions of many PAS proteins may provide DNA binding specificity and stabilize PER-PER interactions suggests that loops within PAS and even linkers between PAS domains and accessory domains, often missing from crystal structures, may confer important functional properties on PAS dimers. Resolving the structure of more loops could help us understand their role in signaling.

7. Unstructured and unsolved regions of the hypervariable regions are important: The known roles of the C-termini in transcriptional activation and repression, as well as ligand-included activation, suggest that this large region of most PAS proteins is likely to interact with PAS and have a significant influence on how these domains function and how these pathways signal.

8. Unstructured regions of PAS may enhance DNA Recognition: While the basic α-helix of the bHLH domain is well known as a primary driver of DNA enhancer sequence recognition, structural data, such as the observation that the Gβ-Hβ loop of HIF2α establishes direct contacts with response elements on the DNA, may be an indication that PAS domains can increase specificity toward genomic targets in many PAS dimers.