Consciousness, neurobiology and quantum mechanics:

The case for a connection

Stuart Hameroff

Departments of Anesthesiology and Psychology

Center for Consciousness Studies

The University of Arizona, Tucson, Arizona

I. Introduction: The problems of consciousness

Consciousness involves phenomenal experience, self-awareness, feelings, choices, control of actions, a model of the world, etc. But what is it? Is consciousness something specific, or merely a byproduct of information processing? Whatever it is, consciousness is a multi-faceted puzzle. Despite enormous strides in behavioral and brain science, essential features of consciousness continue to elude explanation. Unresolved problems include:

1)      Neural correlates of conscious perception apparently occur too late—150 to 500 milliseconds (msec) after impingement on our sense organs—to have causal efficacy in seemingly conscious perceptions and willful actions, often initiated or completed within 100 msec after sensory impingement. For example in the color phi and cutaneous rabbit anomalies, the brain apparently fills in conscious sensory information that is not yet available (Kolers & Grunau 1976, Geldard & Sherrick 1972, c.f. Dennett & Kinsbourne 1992). Preparation of speech can precede conscious identification of heard words to which one is responding (Velmans 1991, Van Petten et al 1999). And in tennis, specific movements to return a fast-moving ball precede conscious identification of ball location and trajectory (McCrone 1999, Gray 2004).[i] Nonetheless, subjectively (i.e. we feel as though) we consciously perceive and respond to these perceptions (e.g. Velmans 1991, Gray 2004, Koch 2004).  

2)      How does the brain provide binding: fusion of a) aspects in one modality (e.g. visual shape, color and motion), b) different modalities (e.g. sight and sound), c) temporal binding of synchronous events sensed asynchronously (e.g. sight and touch) and d) allocentric (simulated external world), egocentric (personal point of view) and enteroceptive (bodily sensation) spaces into unified conscious moments (Gray 2004)?

3)      Electrophysiological correlates of consciousness and attention (e.g. gamma EEG/coherent 40 Hz) may be incompatible with the presumed neural-level correlate of consciousness—trains of axonal action potentials (spikes)—and network-level correlate of consciousness—Hebbian assemblies of axonal-dendritic neurotransmitter-mediated synaptic networks.

4)      The vast majority of brain activity is nonconscious. What distinguishes nonconscious activity from consciousness? 

5)      The hard problem: how does the brain produce qualia, the raw components of phenomenal experience—the smell of a rose, the felt qualities of emotions and the experience of a stream of conscious thought? Why is there conscious experience associated with the brain at all (e.g. Chalmers 1996)?

Prevalent approaches assume that consciousness arises from information processing in the brain, with the level of relevant detail varying among philosophical stances. Generally, all-or-none firings of axonal action potentials (spikes) are seen as the fundamental states, or currency of brain function and equated to roles performed by unitary information states and switches in computers. Consciousness is said to emerge from complex computation: nonlinear dynamics of axonal-dendritic neuronal networks sculpted by modulation of spike-mediated chemical synapses (Hebbian assemblies) form meta-stable patterns—attractors—identified with conscious experience (e.g. Scott 1995, Freeman 2001).

I will refer to all contemporary approaches (perhaps unfairly) as classical functionalism. The implication is that if a robot were precisely constructed to mimic the brain activities which orthodox neuroscience assumes to be relevant to consciousness and perform functions which in a human being are associated with consciousness, then the robot would be conscious regardless of the material from which it was made.

Classical functionalist explanations of the problems stated above are (roughly):

1)      Near-immediate conscious perception and volition are illusions; nonconscious processes initiate many actions (e.g. Velmans 1991, Koch & Crick 2001, Wegner 2002)

2)      Binding—e.g. temporal binding in Dennett’s (1989) multiple drafts model—results from edited memory, rather than real-time unified conscious perception.

3)      Electrophysiological activities measured from scalp, brain surface or within brain extracellular spaces (e.g. gamma EEG/coherent 40 Hz, Section IIId) which seem to correlate with cognition and consciousness are discredited, apparently because axonal spikes fail to show synchronized firing to account for coherence (Shadlen & Movshon 1999, Crick & Koch 2001).

4)      Nonconscious processes compete, with the content of the most active (or optimally synchronized) neuronal groups winning to gain consciousness (e.g. Dennett 1991).

5)      Conscious experience is an emergent property of functional information processing (e.g. Scott 1995, Freeman 2001).  

Consequently, classical functionalism deconstructs consciousness into an out-of-the-loop, illusory set of epiphenomena.[ii] While this might prove true, the view has developed as a default position due to lack of credible alternative and (I will argue) faulty assumptions. Presumed input-output capabilities of individual neurons and neuronal assemblies are tailored to fit the computer analogy, omit essential neurobiological ingredients and miss the target.[iii] Specifically, I will argue that axonal spikes and chemical synaptic transmissions are not the primary currency of consciousness, that electrophysiological correlates of consciousness derive from dendritic activities linked by window-like gap junctions, that glia are involved and that quantum processes in intra-dendritic cytoskeletal microtubules are the actual substrate for consciousness.

Ten years ago Roger Penrose and I put forth a model called orchestrated objective reduction (Orch OR) based on quantum computation in cytoskeletal microtubules inside the brain’s neurons[iv] (Penrose & Hameroff 1995, Hameroff & Penrose 1996a, 1996b, c.f. Hameroff 1998a, 1998b, Woolf & Hameroff 2001). Orch OR has been viewed skeptically by mainstream scientists and philosophers. One apparently valid reason to discount Orch OR is that technological quantum computation is designed to occur in isolation at extremely low temperatures to avoid decoherence—disruption of seemingly fragile quantum states by thermal/environmental interactions. Thus quantum computing at brain temperature in an apparently liquid medium appears impossible. However quantum processes in biological molecules not only occur, but are enhanced at higher temperature (Ouyang & Awschalom 2003). Furthermore the neuronal interior can exist in an isolated, non-liquid gelatinous ordered state (Pollack 2001, Section Vb). Another objection—that quantum states inside one neuron could not extend to others across cellular boundaries—prompted the suggestion that quantum tunneling through window-like gap junctions (which essentially fuse neurons into hyper-neurons, Section IIIe) could enable such extension. Gap junction networks are now shown to be widely prevalent in the brain and to mediate gamma EEG/coherent 40 Hz neuronal activity, the best electrophysiological correlate of consciousness (Section IIId). Finally, Orch OR has been discounted because it differs so markedly from conventional approaches, despite being perfectly consistent with neurobiology. 

In this paper connections among consciousness, neurobiology and quantum mechanics are proposed. They are previewed here:

Consciousness and neurobiology (Section III): The neural correlate of consciousness is in dendrites of cortical neurons interconnected by gap junctions, forming Hebbian ‘hyper-neurons’. Chemical synapses and axonal spikes convey inputs to, and outputs from, conscious processes in hyper-neuron dendrites, consistent with gamma EEG/coherent 40 Hz and the post-synaptic mechanism of general anesthesia. The molecular correlate of consciousness is the intra-dendritic cytoskeleton, specifically microtubules and related proteins whose information processing triggers axonal spikes and regulates synapses.

Neurobiology and quantum mechanics (Section IVc&d): At its core, all chemistry (and biochemistry) is quantum mechanical, though quantum effects are generally considered to wash out at supra-molecular levels due to environmental interactions (decoherence). However in some circumstances biology may utilize quantum effects at mesoscopic and macroscopic scales (e.g. Davies 2004). Specifically, certain proteins act as quantum levers whose functional conformational states are governed by weak quantum forces. Such proteins mediate effects of anesthetic gases which impair the quantum forces, erasing consciousness while sparing other brain activities. Thus only proteins directly involved in consciousness are quantum levers (which can function as quantum bits, or qubits in quantum computation). Evidence suggests that mechanisms have evolved to counter decoherence and enable large scale quantum states in the brain at 37.6 degrees centigrade.       

Quantum mechanics and consciousness (Section Va): The conscious observer has been implicated in quantum mechanics since its inception. Experiments show that quantum superpositions (particles/systems existing in multiple states or locations simultaneously, governed by a quantum wavefunction) persist until measured or observed, then reduce/collapse to definite states and locations. Interpretations vary: in one form of the Copenhagen interpretation the conscious observer causes collapse/reduction of quantum superpositions, placing consciousness outside physics. David Bohm (e.g. Bohm and Hiley 1993) proposed that the wavefunction contains active information which guides the movement of particles, and that consciousness was associated with active information. Like Bohm, the multiple worlds hypothesis (Everett 1957) avoids collapse/reduction but requires an infinity of minds for each individual.[v] Decoherence theory avoids isolated superpositions (and consciousness). Henry Stapp’s view (Stapp 1993) identifies consciousness with collapse/reduction but doesn’t specify a cause of collapse/reduction, or distinction between conscious and nonconscious collapse/reduction. The objective reduction (OR) of Roger Penrose identifies consciousness with collapse/reduction, specifies a cause and threshold, and connects consciousness to fundamental spacetime geometry, introducing mechanisms for non-computable Platonic influences and proto-conscious qualia. And like Stapp’s view, Penrose OR connects to Whitehead’s philosophical approach to consciousness.

We begin with a consideration of the timing of conscious experience.

II. Time and consciousness

a. Is consciousness continuous or a sequence of discrete events?

William James (1890) initially considered consciousness as a sequence of specious moments but then embraced a continuous stream of consciousness. Alfred North Whitehead (1929, 1933) portrayed consciousness as a sequence of discrete events: occasions of experience. As motion pictures—in which sequential frames are perceived as continuous—became increasingly popular, so did the notion of consciousness as discrete events, e.g. the perceptual moment theory of Stroud (1956). Evidence in recent years suggests periodicities for perception and reaction times in the range of 20 to 50 msec (gamma EEG) and another in the range of hundreds of msec (alpha and theta EEG), the latter consistent with saccades and the visual gestalt (VanRullen & Koch 2003, VanRullen & Thorpe 2001). Based on a proposal for memory by Lisman and Idiart (1995), VanRullen and Koch (2003) suggested a multiplex for visual perception in which a series of fast gamma waves (each corresponding to specific components of vision) rides on a slower, e.g. theta wave (corresponding to an integrated visual perception). A similar, previous model of gamma/theta complex waves supporting quantum mechanisms underlying conscious vision (Woolf & Hameroff 2001) will be discussed in Section VIIIa.  Freeman (2004a) has shown cinematographic effects in neural excitations in the brain, supporting the notion of discrete conscious frames.

If consciousness is a sequence of events, what is its rate or frequency? Can it vary? In the midst of a car accident, victims often report that time seems to slow down. Does this excited state involve an actual increase in the rate of subjective conscious moments per objective time? What are conscious moments, why are they subjective and how do they relate to neurobiology?

b. The timing of conscious experience

Many behaviors apparently happen too quickly to be initiated by consciousness. Max Velmans (1991) lists examples: analysis of sensory inputs and their emotional content, phonological and semantic analysis of heard speech and preparation of one’s own spoken words and sentences, learning and formation of memories, and choice, planning and execution of voluntary acts. Consequently, subjective feeling of conscious control of these behaviors is deemed illusory (Wegner 2002). 

In speech, evoked potentials indicating conscious word recognition occur at about 400 msec after auditory input, however semantic meaning is appreciated (and response initiated) after only 200 msec. As Velmans points out, only two phonemes are heard by 200 msec, and an average of 87 words share their first two phonemes. Even when contextual effects are considered, semantic processing and initiation of response occurs before conscious recognition (Van Petten et al 1999). 

Jeffrey Gray (2004) observes that in tennis “The speed of the ball after a serve is so great, and the distance over which it has to travel so short, that the player who receives the serve must strike it back before he has had time consciously to see the ball leave the server’s racket. Conscious awareness comes too late to affect his stroke”. John McCrone (1999): “[for] tennis players…facing a fast serve…even if awareness were actually instant, it would still not be fast enough....”.

Visual recognition of an object’s shape, color, motion and semantic meaning occur in different parts of visual cortex, and at different times (Zeki and Bartels 1998, Zeki 2003). Yet we consciously perceive these features simultaneously (the temporal binding problem).

Touch also involves temporal binding. If you tap your foot with your finger, the foot and finger sensations occur simultaneously. Yet the sensory signal from your foot requires significantly longer to reach sensory cortex than does that from your finger. How does the brain provide synchrony?

In the cutaneous rabbit experiment (Geldard and Sherrick 1972, 1986) a subject’s arm is mechanically “tapped” at three locations along the arm, e.g. 5 taps at the wrist followed by 2 at the elbow then 3 more on the upper arm. However subjects report a regular sequence of taps traveling in equidistant increments, as if a small animal were hopping along their arm. The “departure” from the wrist begins with the second tap, yet if the upper taps are not delivered, all 5 wrist taps are felt at the wrist. It is as if the brain knows in advance there will be (or not be) taps further along the arm.

Figure 1. The “color phi” phenomenon (Kolers & von Grunau 1976). Top left: an observer views a screen on which a red circle appears on the left, disappears, and then a green circle appears on the right. Bottom left: the observer’s conscious (reported) experience is of a red circle moving from left to right, changing to green halfway across. Upper right: the retrospective construction explanation is that the observer’s real time perception is of two separate circles, subsequently revised and recorded in (delayed) memory as the red circle moving and changing to green halfway across. Bottom right: Quantum explanation in which the brain sends subconscious quantum information backward in time, filling in the red circle changing to green halfway across.  

In the “color phi” effect (Kolers and von Grunau 1976) a red spot appears briefly on the left side of a screen, followed after a pause by a green spot on the right side. Observers report one spot moving back and forth, changing color halfway across (Figure 1). Does the brain know in advance to which color the dot will change?

Perhaps the most perplexing experiments regarding time and mental events were done by Benjamin Libet and colleagues in the 1960s and 1970s. They studied awake, cooperative patients undergoing brain surgery with local anesthesia so that the patients’ brains were exposed (e.g. Libet et al 1964, 1979, Libet 2004). In these patients Libet was able to access, identify, record from and stimulate specific areas of somatosensory cortex (postcentral gyrus) corresponding to the skin of each patient’s hand (Figure 2). He found that direct electrical stimulation of the somatosensory ‘hand’ area of cortex resulted in brain electrical activity (DCR: direct cortical response due to neuronal dendritic activity). This in turn caused conscious sensation referred to the hand, but only after a train of threshold-level pulses (and DCR activity) lasting about 500 msec. This requirement of ongoing, prolonged electrical activity from direct cortical stimulation to produce conscious experience (‘Libet’s 500 msec’) was confirmed by Amassian et al (1991), Ray et al (1999), Pollen (2004) and others.

But what about normal sensory perception? Single, threshold-level stimuli to the hand or elsewhere are seemingly perceived consciously almost immediately; no 500 msec delay occurs when we touch something. In the brain somatosensory cortex, threshold level stimuli at the skin of the hand cause a primary evoked potential (EP) 10 to 30 msec after skin stimulation, followed by ongoing activity of several hundreds of msec, very much like Libet’s DCR.

But the primary EP is not sufficient for conscious experience:

• ·        A single stimulus delivered to subcortical brain regions in the sensory pathway causes a primary EP without conscious experience or prolonged activity (Libet et al 1967, Jasper and Bertrand 1966).[vi]
• ·        Sub-threshold skin stimulation causes a primary EP, but no prolonged cortical activity nor conscious experience.
• ·        Under general anesthesia, skin stimulation of any kind can cause a primary EP but no ongoing cortical activity nor conscious experience.

On the contrary, prolonged cortical activity (‘Libet’s 500 msec’) is both necessary and sufficient for conscious experience, but in the absence of a primary EP produces only delayed conscious experience.

• ·        Libet’s DCR patients (also Amassian et al 1991, Ray et al 1999) had ~500 msec delayed conscious experience of skin stimulation without a primary EP caused by a train of pulses delivered to cortex. Pollen (2004) showed a similar delay with visual phosphenes after occipital cortex stimulation.

Libet’s conclusion was that the 500 msec prolonged cortical activity is the sine qua non for conscious experience—the NCC, or neural correlate of consciousness. The primary EP is necessary (but not sufficient) for near-immediate conscious experience.[vii],[viii] Primary EP and prolonged activity together produce near-immediate conscious experience.

Figure 2. Libet’s experiments and explanation (Libet et al 1979, Libet 2004). Patient (left) was accessed 1) at hand area of somatosensory cortex, and 2) skin of corresponding hand. Top: Direct cortical stimulation of electrical pulses every 50 msec caused cortical brain activity which was required to proceed for 500 msec to cause conscious experience of a sensation in the hand. Middle: Single pulse to the skin of the hand caused primary evoked potential (EP) after 10 to 30 msec and ongoing brain activity for at least 500 msec. Conscious experience occurred concomitant with primary EP. Bottom: Libet’s explanation—500 msec ongoing activity required for neuronal adequacy which is referred backward in time to the primary EP.     

But if the neural correlate of conscious experience is delayed for 500 msec, how/why do we seem to perceive sensory events almost immediately? Are we living in the past, but remembering (falsely) being in the here and now, as Dennett suggests (next Section)? To address the question, Libet and colleagues proposed and tested a rather outrageous hypothesis—that the perception of a stimulus was indeed delayed for 500 msec of brain activity but subjectively referred backward in time to the primary evoked potential 10 to 30 msec after stimulus. 

Experiments were performed in which patients received both direct cortical stimulation of the hand area and stimulation of the actual skin of the hand. Although both were perceived in the hand, the two were qualitatively different so the subjects could distinguish them. Stimulation of the two sites were given in close, but varying temporal proximity (i.e. within one second), and the patients asked which stimulus was felt first.

The patients reported that the sensations generated at the skin appeared before the cortically induced sensation, even when the skin pulse was delayed by some hundreds of msec after the start of the cortical stimulation. Only when the skin pulse was delayed for about 500 msec after the cortical stimulation did the subjects report feeling the two stimuli simultaneously. The skin-induced experience appeared to have no delay. The cortically-induced experience was delayed 500 msec relative to the skin-induced sensation.

So both skin-induced and cortically-induced sensations required 500 msec of cortical processing, but the skin-induced sensation was experienced almost immediately. Unlike the cortically induced experience the skin-induced sensation was marked by a primary EP. Was that the difference?

To investigate this question, Libet also studied patients with electrodes implanted (for therapeutic purposes) in the medial lemniscus below the thalamus, i.e. in the brain’s sensory pathway enroute from hand to cortex. He determined that stimulation of the medial lemniscus could produce a conscious experience only after 500 msec of stimulation and cortical activity. But unlike direct cortical stimulation (and like skin stimulation) medial lemniscus stimulation produced primary EPs. Libet and his colleagues then performed another set of experiments comparing stimulation of the hand with stimulation of medial lemniscus, coupling the two stimuli at varying time intervals. They found no delay of the medial lemniscus stimuli compared to skin stimuli.

But the patients felt nothing if medial lemniscus stimulation was interrupted prior to the full 500 msec stimulation. So prolonged cortical activity was necessary for conscious experience and the primary EP was necessary for near-immediate subjective experience.     

Libet came to the following conclusions:

• ·        Conscious perception requires brain activity for ~500 msec to achieve neuronal adequacy. 
• ·        Information is referred up to 500 msec backward in time to the primary evoked potential—10 to 30 msec after peripheral stimulation—for near-immediate conscious perception.

Libet’s results and conclusions have been repeatedly challenged but never refuted (Libet 2002, 2003).

c. Taking backward time referral seriously

How do we resolve these temporal anomalies? The color phi effects apparently “…leave us a choice between a retrospective construction theory and a belief in clairvoyance” (Goodman 1978).

Daniel Dennett (1991, c.f. Dennett & Kinsbourne 1992) chose retrospective construction in the context of a multiple drafts model in which sensory inputs and cognitive processing produce tentative contents under continual revision. A definitive, final edition is inserted into memory, overriding previous drafts. A key feature is that consciousness (e.g. of a particular perception) occurs not at any one specific moment, but arbitrarily in time, like the onset of fame, or end of a war. The brain retrospectively creates content or judgment, e.g. of intervening movement in the color phi experiment.[ix] 

According to retrospective construction (I presume): 1) tennis players see and hit balls unconsciously, but remember seeing and hitting consciously.[x] 2) Sensory components of objects or events are perceived asynchronously but remembered as being synchronous. 3) In the cutaneous rabbit experiment, the subjects feel wrist taps, then elbow taps, then upper arm taps, but remember a sequence of evenly spaced taps. 4) In the color phi phenomenon the observer sees the left side red spot, then the right side green spot, but remembers the red spot moving and changing colors mid-stream.

Thus according to Dennett and many others, smooth, real-time conscious experience is an edited construction—an illusion. Dennett and Kinsbourne (1992) have a more difficult time dispensing with Libet’s findings, describing them as “interesting but inconclusive.”

Libet performed other experiments related to volition. Kornhuber and Deecke (1965) had recorded over pre-motor cortex in subjects who were asked to move their finger randomly, at no prescribed time. They found that electrical activity preceded finger movement by 800 msec, calling this activity the readiness potential. Libet and colleagues (1983) repeated the experiment except they also asked subjects to note precisely when they consciously decided to move their finger. This decision came approximately 200 msec before movement, hundreds of msec after onset of the readiness potential. Libet concluded that many seemingly conscious actions are initiated by nonconscious processes.   

Libet didn’t consider backwards referral in volition because antedating in his sensory experiments was pinned to the primary sensory EP, and no such marker existed in the spontaneous finger movement experiments. However voluntary acts in response to stimuli (hitting a ball, choosing a word in a sentence) do have such markers, as would binding of temporally asynchronous perceptual components of synchronous events. Nor did Libet consider backward referral as implying an actual reversal in time, but a phenomenon akin to retrospective construction. Libet (2000, p. 7) says:

“…the timing of a sensation is subjectively referred….not that the conscious sensation itself jumped backwards in time…the content of the subjective experience…is modified by the referral to the earlier timing signal.”

But consciousness lagging a half second behind reality would render it largely epiphenomenal (and illusory).[xi] We would be (in the words of T.H. Huxley) “helpless spectators”. Perception would be a jangle of disconnected events edited for memory, too late for conscious control of many seemingly conscious actions. Perhaps so, but is there a possible alternative?

Yes. To account for Libet’s results, Roger Penrose (1989, c.f. Wolf 1989) suggested that the brain sends unconscious quantum information backward through time. In the quantum world, time is symmetrical, or bi-directional (as it also appears to be in unconscious dreams—Section VI).[xii] Aharonov and Vaidman (1990) proposed that quantum state reductions send quantum information backward in time; such backward time referral is the only apparent explanation for experimentally observed EPR effects in quantum entanglement (Figure 3, Section Va, Penrose 2004, c.f. Bennett and Wiesner 1992). One could say the quantum world is timeless, or has no flow of time.

Figure 3. Backward time in the EPR effect. A. The Einstein-Podolsky-Rosen (EPR) experiment verified by Aspect et al (1982), Tittel et al (1998) and many others. On the left is an isolated entangled pair of superpositioned complementary quantum particles, e.g. two electrons in spin up and spin down states. The pair is separated and sent (through environment but unmeasured) to different locations/measuring devices kilometers apart. The single electron at the top (in superposition of both spin up and spin down states) is measured, and reduces to a single classical state (e.g. spin down). Instantaneously its complementary twin kilometers away reduces to the complementary state of spin up (or vice versa). The effect is instantaneous over significant distance, hence appears to be transmitted faster than the speed of light. B. The explanation according to Penrose (2004, c.f. Bennett and Wiesner 1992) is that measurement/reduction of the electron at the top sends quantum information backward in time to the origin of the unified entanglement, then forward to the twin electron. No other reasonable explanation has been put forth. 

Quantum information cannot actually convey information, and is thus a misnomer (Penrose now calls it ‘quanglement’ because of its role in quantum entanglement). Quanglement can only modify classical information, but mere modification is highly significant in EPR experiments and quantum technology (Section V). Quantum information/quanglement going backward in classical time is also constrained by possible causality violations, i.e. causing an observable change resulting in a paradox like going back in time to kill your ancestor, thereby preventing your birth. Any effect which could be even possibly measured or observed may be prohibited. Nonconscious backward referral of quantum information/quanglement which modifies existing information in the brain (e.g. adding qualia to primary evoked potentials, influencing choices) would not violate causality because the effects are unobservable before they occur.[xiii]

Backward time referral of unconscious quantum information/quanglement in the brain could provide temporal binding and near-immediate perception and volition, rescuing consciousness from illusory epiphenomenon (i.e. enabling near-immediate conscious decisions based on sensory information referred from the near future). How this could actually happen will be discussed in Section VII, but we next turn to where it could happen—the neural correlate of consciousness.

III. The neural correlate of consciousness

a. Functional organization of the brain

Most brain activities are nonconscious; consciousness is a mere “tip of the iceberg” of neural functions. Many brain activities—e.g. brainstem-mediated autonomic functions—never enter consciousness. While consciousness is erased during general anesthesia, nonconscious brain EEG and evoked potentials continue, although reduced.[xiv]

Functional units corresponding to particular mental states are generally considered as  networks or assemblies of neurons, originally described by Donald Hebb (1949, see also Scott 2004). Hebb described assemblies as closed causal loops of neurons which could be ignited by particular inputs and remain active for hundreds of msec, following which another related assembly would ignite, then another and so on in a phase sequence. Hebb described assemblies as “three dimensional fishnets” of many thousands of neurons. At any one time a single particular assembly would be the neural correlate of consciousness (NCC).

Why would a particular assembly be conscious? Dennett’s multiple drafts model proposes, as does Susan Greenfield’s (2000) epicenter model, that brain activity accompanying consciousness is the same in kind as unconscious brain activity, except more so. Regardless of location, if activity of a neural assembly representing a specific set of content exceeds all other in some type of competition, it takes the prize of entering into consciousness.

What the precise neural activity accompanying consciousness actually is remains to be seen, but where does it occur? Global workspace theory describes multiple specialized brain areas interconnected in a coordinated, though variable manner. Bernie Baars (1988) introduced the concept which was elaborated anatomically by Changeux and Dehaene (1989, see also Dehaene & Naccache 2001). Crick and Koch (1990), and Edelman and Tononi (2000) have similar approaches.

The basic idea is that consciousness occurs primarily in a horizontal layer of interconnected cortical neurons sandwiched between ascending, bottom-up inputs from thalamus and basal forebrain, and top-down executive functions from pre-frontal cortex.[xv] Bottom-up inputs convey sensory information, as well as general arousal and highlighted saliency such as emotional context from basal forebrain inputs (Woolf 1997, 1999). Top-down influences categorize and manipulate unexpected features (Koch 2004), e.g. those associated with danger, reward etc. Acting together, bottom-up and top-down activations select a neural assembly—a specific subset of cortical-cortical projections—for attention and consciousness, prompting sufficient activity for the assembly to become the NCC. Over time, the NCC and its contents change with dynamically shifting, temporary alliances of neurons and assembly makeup. Global workspace models demonstrate a functional architecture which could accommodate consciousness.

Placing consciousness between bottom-up and top-down neuronal pathways agrees with Ray Jackendoff’s (1987) intermediate level theory which notes we are not normally aware of pure sensation, nor of pure conceptual structures, but an optimized admixture of the two. The intermediate level is also consistent with Jeffrey Gray’s (1995, 1998) comparator hypothesis in which consciousness is the output of a process which compares available (e.g. incoming, bottom up) information against anticipatory (executive, top down) schemata.  

Evidence from vision supports both Jackendoff’s contention and global workspace theory. Visual inputs synapse in thalamus and project raw data mostly to primary visual area V1 in the posterior occipital cortex. V1 then sends information forward to other regions of visual cortex[xvi] e.g. V2 where shape and contour are recognized, V4 where color is perceived and V5 where motion is detected. These and other secondary visual areas project to pre-frontal cortex for categorization and planning. Pre-frontal cortex then projects back towards V1 and other visual areas. Crick and Koch (2001) have argued the NCC of vision lies not in V1 or pre-frontal cortex but in intermediate areas. In Jackendoff’s terms, V1 houses “pure sensation unaffected by conceptual interpretation”. Visual consciousness occurs in the middle—shifting assemblies of cortical-cortical projections sandwiched between (but possibly including) V1 and pre-frontal cortex.

However Zeki (1999) has shown that excessive activity in any feature-selective region may be sufficient on its own for that feature to enter consciousness. Thus activity in V4 alone can result in the experience of color.

Other NCC candidates candidates include the hippocampus in Jeffrey Gray’s comparator hypothesis, and the brainstem in Antonio Damasio’s (1999) and Jaak Panksepp’s (1999) separate views of emotional core consciousness. Thus while consciousness occurs generally in what is termed a global workspace, it may also arise in more localized and perhaps separate regions. The question remains how/why consciousness arises in any region. What aspect of neural activity gives rise to consciousness?     

b. Cerebral cortex and neuronal assemblies

Cerebral cortex is hierarchical in two different ways (Koch 2004). Microscopically, layer 4  receives primary sensory inputs from thalamus and is thus on the bottom. Geography aside, layers 1-3 and 6 are more or less in the middle. Layer 5 giant pyramidal cells (which convey the verdicts of cortical processing to subcortical regions) are at the top of the hierarchy. This arrangement is nested in a larger scale anatomical hierarchy with primary sensory areas (such as V1 for vision) at the bottom, and pre-frontal executive cortex at the top. Consistent with Jackendoff’s intermediate theory, shifting assemblies of many types of neurons sandwiched throughout numerous cortical regions appear to act as the NCC.

Particular Hebbian assemblies may be formed and strengthened primarily by alterations in dendritic morphology leading to enhanced synaptic activity and lowered threshold for specific circuits. Assemblies sculpted by post-synaptic changes—synaptic plasticity—are the cornerstone of theoretical mechanisms for learning, memory and the NCC. The mechanisms of plasticity include altered number, sensitivity and clustering of post-synaptic receptors, optimal geometry of dendritic spines and branchings, dendro-dendritic connections, and changes in decremental conductance of post-synaptic potentials (e.g. Hausser et al 2000). All these changes are mediated by structures within neuronal dendritic interiors, namely the cytoskeleton (e.g. Dayhoff et al 1994).

c. Axons and dendrites

Since Cajal, the neuron doctrine has been that information flows from an incoming axon across a chemical synapse to a dendrite or cell body of another neuron. When a post-synaptic threshold is met from accumulation of excitations (offset by inhibitions), the second neuron’s axon fires—an action potential or spike is triggered at the proximal axon hillock. Mediated by fluxes of sodium ions across membrane channels, spikes propagate along the axon to reach another synapse where they influence the release of neurotransmitters. Each neuron has only one axon, though they may branch downstream. Thus multiple post-synaptic inputs are integrated to lead to one output, the all-or-none firing of a spike.[xvii]

Spikes can be measured and quantified by electrodes which traverse or pass near axonal membranes. Thus we know that spike frequency (and possibly patterns) correlates with intensity of stimulus and/or behavior (e.g. Britten et al 1992). Spikes travel rapidly and are robust, not degrading even when conveyed over long distances. They are widely assumed to be the primary means of signaling and information transfer in the brain, and thus the currency or substrate—the neural code—of consciousness. The notion of multiple inputs integrated to a threshold leading to a single output lends itself well to computer models and analogies. Spike = bit! 

However there are other cellular-level candidates for the NCC. Electrodes on the scalp or brain surface detect mostly dendritic dipole potentials from pyramidal cells with axial symmetry, i.e. oriented perpendicular to the brain surface (Freeman 2001). Electrodes implanted into the brain detect mainly local field potentials (LFPs) generated from cortical interneurons with radial symmetry, linked mostly by dendro-dendritic gap junctions and inhibitory chemical synapses. Thus synchrony in the EEG and LFPs derive not from axonal spikes but from dendritic activities. Moreover the BOLD signal used in fMRI, widely assumed to represent neural metabolic activity related to consciousness, correspond more closely with LFPs than axonal spikes (Logothetis 2002).

Some have argued (e.g. Libet 2004, McFadden 2000, Pockett 2000) that the brain’s complex electromagnetic field (manifest as global LFPs and surface potentials) constitutes the NCC. However as Koch (2004) points out, the brain’s electromagnetic field per se is a crude and inefficient means of communication.

On the other hand, the dendritic activities that generate LFPs and/or surface potentials may indeed best represent the NCC. Eccles (1992) as well as Pribram (1991) suggested that dendrites host consciousness, with axonal spikes conveying the outputs of consciousness to other neurons, brain regions and initiating motor responses.

Neurotransmitter binding at synaptic receptors changes voltage potentials across dendritic or cell body membrane, causing either excitatory or inhibitory post-synaptic potentials (IPSPs, EPSPs) and in some cases dendritic action potentials (Buzsáki and Kandel 1998, Schwindt and Crill 1998). These are then presumed to summate as membrane potentials to reach threshold for spike initiation at the proximal axon hillock.

Is integration to trigger spikes the full extent of dendritic function? Some cortical neurons have no axons, dendrites interact with other dendrites (e.g. Isaacson and Strowbridge 1998, Sassoè-Pognetto and Ottersen 2000) and extensive dendritic activity may occur without causing spikes. Dendritic membrane fluctuations below spike threshold (generally considered noise by neuroscientists) may oscillate coherently across wide regions of brain (Arieli et al 1996, Ferster 1996).

Highly branched dendritic structure is capable of information processing beyond summation of membrane potentials. Evidence shows complex logic functions in local dendritic compartments, signal boosting (e.g. at branch points), filtering and changing axon hillock sensitivity (Sourdet and Dehanne 1999, Poirazi and Mel 2001, Shepherd 1996, 2001).

Figure 4. Characterizing neurons. Left: Illustration of an actual pyramidal neuron with multiple apical and basilar dendrites (top and middle) and a single axon heading downward. Two incoming axons are shown synapsing on apical dendrites. Middle: A cartoon neuron as depicted in neural network and functionalist models. Two incoming axons are shown synapsing on the cell body/dendrite. Right: A cartoon neuron as utilized in this paper, showing three dendrites, cell body and a single axon heading downward. The internal cytoskeleton—microtubules interconnected by microtubule-associated proteins— is shown schematically; in dendrites and cell body the microtubules are short, interrupted (and of mixed polarity, not visibly apparent). In the axon the microtubules are continuous (and of uniform polarity, not visibly apparent). Two incoming axons synapse on dendritic spines.   

Nor is dendritic processing necessarily limited to membrane potentials. Many post-synaptic receptors are metabotropic, sending signals internally into the dendritic cytoskeleton, activating enzymes,[xviii] causing conformational signaling and ionic fluxes along actin filaments and dephosphorylating microtubule-associated protein 2 (MAP2) which links microtubules into cytoskeletal networks. MAP2 activity is necessary for learning and memory, and is the largest consumer of dendritic metabolic energy (Theurkauf and Vallee 1983, Aoki and Siekevitz 1988, Johnson and Jope 1992). Changes in the cytoskeleton regulate synaptic plasticity (Halpain and Greengard 1990, Van der Zee et al 1994, Woolf 1998, O'Connell et al 1997 Whatley and Harris 1996, Woolf 1998, Woolf et al 1999, Fischer et al 2000, Sanchez et al 2000, Matus 2000, Khuchua et al 2003).

Dendritic processing is assumed to be constrained by global all-or-none output through the axon, and to exist merely to trigger axonal spikes. But neither assumption is  substantiated. The full extent of dendritic internal processing is unknown but its capabilities are enormous. For example synaptic activity causes glycolytic production of ATP in dendritic spines, energy which may be used for ion channels as well as protein synthesis and signal transduction into the dendritic cytoskeleton (Wu et al 1997, Siekevitz 2004). Kasischke and Webb (2004) suggested that brain function might be “…. more refined on a higher temporal and smaller spatial scale.”

Figure 4 shows 1) an actual pyramidal neuron with multiple dendrites; two incoming axons synapse on two different dendrites (a pyramidal neuron is likely to have many thousands of such incoming synapses), 2) a cartoon neuron with two axonal inputs synapsing on a cell body (as presumed in functionalist models), and 3) a more elaborate cartoon neuron with three dendrites (and two incoming synapses) showing the internal cytoskeleton. Figure 5 shows this type of cartoon neuron with a chemical synapse and dendritic-dendritic gap junction.

Figure 5. Cartoon neuron with two types of connections. Internal structure represents nucleus (dark circle) and cytoskeletal microtubules (MTs) connected by strut-like microtubule-associated proteins (MAPs). MTs in axons are continuous (and unipolar) whereas dendritic MTs are interrupted (and of mixed polarity.  Lower left: An incoming axon forms a chemical synapse on a dendritic spine. Close up shows neurotransmitter vesicles in pre-synaptic axon terminal, and post-synaptic receptors on spine connected to intra-spine actin filaments which link to MTs. Upper left: Dendritic-dendritic gap junction is a window between the two neurons. Both the membranes and cytoplasmic interiors of the two cells are continuous.

d. Neural synchrony

The most active neuronal assembly is assumed to undergo a transition from nonconscious representation to consciousness. Why the most active processes should be conscious is the hard problem. But regardless, what exactly is the relevant neural activity? Axonal spike frequency is assumed to be the critical function but depends entirely on dendritic activities. Evidence supports a correlation between consciousness and synchronous activity.

Electrical recording from scalp, brain surface or implanted electrodes reveal synchronous activity at various frequencies of the electroencephalogram (EEG) due to LFPs or surface potentials. Among these, the so-called gamma frequency range between 30 and 70 Hz correlates best with attention and consciousness. Gray and Singer (1989, c.f. Gray et al 1989) found coherent gamma oscillations in LFPs of cat visual cortex which strongly depended on specific visual stimulation. Though the synchrony could occur in a range between 30 and 70 Hz, the coupling phenomenon became known euphemistically as coherent 40 Hz.

Following a suggestion by von der Malsburg (1981) that synchronous neural excitations could solve the binding problem, von der Malsburg and Singer (1988), Crick and Koch (1990),[xix] Varela (1995) and others proposed that the neural correlate of any particular conscious content was an assembly of neurons excited coherently at 40 Hz or thereabouts. Varela (1995) succinctly observed that neural synchrony operated whenever component processes subserved by spatially separate brain regions were integrated into consciousness.

Neural synchrony in the gamma frequency range has been observed in many animal studies using multi-unit scalp, surface and implanted electrodes. They demonstrate synchrony within and across cortical areas, hemispheres and sensory/motor modalities which reflects perceptual gestalt criteria and performance (for review: Singer & Gray 1995, Singer 1999).  Among a smaller number of human studies using scalp EEG and MEG, most have supported a role for synchrony in integration and binding (Joliot et al 1994, Singer 1999, Varela et al 2001, Trujillo et al 2004). Gamma synchrony has been shown to correlate with the perception of sound and linguistic stimuli (Miltner et al 1999, Pantev 1995, Ribary et al 1991), REM dream states (Llinas & Ribary 1993), attention (Fries et al 2002, Tiitinen et al 1993), working memory (Tallon-Baudry et al 1996, 1997), face recognition (Mouchetant-Rostaing et al 2000), somatic perception (Desmedt & Tomberg 1994) and binding of visual elements into unitary percepts, with the magnitude of synchrony diminishing with stimulus repetition (Gruber & Muller 2002). And loss of consciousness associated with onset of general anesthesia is characterized by a decrease in gamma EEG activity which returns when patients awaken (John 2001).[xx]

Figure 6. Neural network/Hebbian assembly of cartoon neurons linked by axonal-dendritic chemical synapses. Information/excitation flows unidirectionally (counter-clockwise) from axon to dendrite through the network. Electrical recordings at various points show single voltage spike potential propagating through the network.

Some human studies have failed to support neural synchrony in perception and cognition. Menon et al (1996) found gamma synchrony restricted to less than 2 cm regions of cortical surface, arguing against long-range coherence. However the study only examined a 7 cm x 7 cm region and other studies show that synchrony drops off at intermediate ranges but then reappears at long range distances (Nunez et al 1997). Some discrepancies  have ensued from differences in methodology (Trujillo et al, 2004). Overall, synchronous gamma/coherent 40 Hz is the best electrophysiological correlate of consciousness.

How is gamma synchrony mediated? Coherence over large distances, in some cases multiple cortical areas and both cerebral hemispheres, shows zero, or near-zero phase lag. Significant phase lags would be expected from the speed of axonal conduction and delays in synaptic transmission (Traub et al 1996).

There is no evidence to support coordinated axonal spiking as the source of gamma synchrony. As Koch (2004) states: “Gamma oscillations can be routinely observed in the local field potential and, less frequently, when recording multi-neuron activity (that is, the summed spikes of neighboring cells). Detecting these rhythms in the spiking patterns of individual neurons has proven to be more problematic….”.

A critical review (Shadlen and Movshon 1999) rejects the relevance of synchrony to temporal binding (and consciousness) based on the lack of coherence of spike activity, perhaps throwing away the baby with the bathwater. However many studies have shown gamma frequency synchronized by dendritic gap junction electrical synapses. Measuring both spikes and dendritic LFPs in multiple regions of cat visual cortex, Fries et al (2002) showed that visual recognition corresponded with gamma frequency EEG emanating from LFPs, not with spikes.

Figure 6 shows a cartoon neuronal network based on axonal spikes and chemical synapses. Excitation/information flows through the network; there is no coherence. Figure 7 shows a gap junction-linked neuronal network (a hyper-neuron, including glial cells) with continuous membrane and cytoplasm. Dendritic membrane throughout the hyper-neuron is excited coherently.

Figure 7. Neural network/Hebbian assembly (‘hyper-neuron’) linked by window-like gap junctions, mostly dendritic-dendritic but also by glial cell gap junctions. Inputs to the hyper-neuron are from axonal-dendritic chemical synapses. Outputs from the hyper-neuron are from axons of hyper-neuron components. Because gap junction-connected neurons depolarize synchronously like “one giant neuron”, electrical recordings at various points show synchronous voltage depolarizations, e.g. at coherent 40 Hz. Both membranes and cytoplasmic interiors are continuous throughout the hyper-neuron.   

 e. Gap junction assemblies—‘hyper-neurons’

Gap junctions, or electrical synapses, are direct open windows between adjacent cells formed by paired collars consisting of a class of proteins called connexins (Herve 2004, Rouach et al 2002). Gap junctions occur between neuronal dendrites, between axons and axons, between neurons and glia, between glia, and between axons and dendrites—bypassing chemical synapses (Traub et al 2001, Froes & Menezes 2002, Traub et al 2002, Bezzi & Volterra 2001). Ions, nutrients and other material pass through the open gaps, so gap junction-connected neurons have both continuous membrane surfaces and continuous cytoplasmic interiors. Membrane depolarizations travel bidirectionally across gap junctions, so neurons connected by gap junctions are electrically coupled, depolarize synchronously  and “behave like one giant neuron” (Kandel et al 2000).

In early development gap junctions link pyramidal cells with each other, with non-pyramidal neurons, and with glia during formation of cortical circuits (Bittman et al 2002). The number of cortical gap junctions then declines so gap junctions were considered irrelevant to cognition or consciousness. However many studies show that gap junctions persist significantly in the adult mammalian brain. Moreover, gap junction circuits of cortical interneurons in adult brains mediate gamma EEG/coherent 40 Hz and other synchronous activity (Dermietzel 1998, Draguhn et al 1998, Hormuzdi et al 2004, Bennett and Zukin 2004, Lebeau et al 2003, Friedman and Strowbridge 2003, Buhl et al 2003, Rozental et al 2000, Perez-Velazquez and Carlen 2000, Galaretta and Hestrin 1999, Gibson et al 1999).

At least ten different connexins are found in mammalian brain, and their placement and function are dynamic (Bruzzone et al 2003, Bennett and Zukin 2004). A single neuron may have numerous gap junction connections, only some of which are open at any one time, with rapid openings and closings regulated by cytoskeletal microtubules, and/or phosphorylation via G-protein metabotropic receptor activity (Hatton 1998). Thus gap junction networks are at least as dynamic and mutable as those crafted by chemical synapses, and may include glial cells (Froes et al 1999). They fulfill the criteria for Hebbian assemblies with the added advantage of synchronous excitations. Networks of gap junction-linked neurons (and glia) have been termed hyper-neurons (John et al 1986).[xxi]

Cortical inhibitory interneurons are particularly studded with gap junctions, potentially connecting each cell to 20 to 50 others (Amitai et al 2002). Many have dual synapses—their axons form inhibitory GABA chemical synapses on another interneuron’s dendrite, while the same two cells share dendro-dendritic gap junctions (Tamas et al 2000, Fukuda and Kosaka 2000, Galaretta and Hestrin 2001). Within each cortical hemisphere there is no apparent limit to the extent of interneuron gap junction networks—hyper-neurons—in which they may form a “large, continuous syncytium” (Amitai et al 2002).

The case for gap junction hyper-neurons involving primary neurons such as pyramidal cells in mature brains, and extending to both hemispheres is less clear. However Venance et al (2000) showed gap junctions between interneurons and excitatory neurons in juvenile rat brain. Pyramidal cells in hippocampal slices show axo-axonal gap junction coupling (Traub et al 2002), and glial cells envelope both axons and dendrites in many chemical synapses. Neuron-glia-neuron gap junctions could thus provide chemical synapses with alter egos as links in hyper-neurons. Thalamo-cortical cells generating synchronous alpha and theta cortical activity are linked by gap junctions in thalamus (Hughes et al 2004), so thalamo-cortical projections (or trans-corpus callosum pathways) could couple both hemispheres in hyper-neurons to account for bilateral synchrony.

In principle, all the brain’s neurons and glia could be linked together by gap junctions. However too many active gap junctions and near total synchrony (e.g. as in seizures) would reduce the brain’s information processing capacity. More than three active gap junctions per neuron (i.e. with three different neurons or glia) would connect the entire brain into a single hyper-neuron topology.[xxii] Thus pruning and sparseness are necessary. For the purpose of this paper, hyper-neurons will imply gap junction-linked cortical interneurons, glia, primary cortical neurons such as pyramidal cells and perhaps others such as thalamo-cortical neurons which can extend throughout both cerebral hemispheres and subcortical areas.

Brain-wide gamma synchrony mediated by gap junctions is the best electrophysiological NCC. A logical conclusion is that gap

junction networks—hyper-neurons—are the cellular-level NCC. Can that help explain consciousness?

A key feature of gap junction hyper-neurons is continuous dendritic membranes which depolarize coherently. Another key feature is continuous cytoplasmic interiors. 

f. Neuronal interiors and the cytoskeleton

The neuronal interior is the next NCC frontier. Membrane-based neuronal input-output activities involve changes in synaptic plasticity, ion conductance, neurotransmitter vesicle transport/secretion and gap junction regulation—all controlled by the intra-neuronal networks of filamentous protein polymers known as the cytoskeleton. If simple input-output activities fully described neural function, then fine-grained details might not matter. But simple input-output activities—in which neurons function as switches—are only a guess, and most likely a poor imitation of the neuron’s actual activities and capabilities.

To gauge how single neuron functions may exceed simple input-output activities, consider the single cell organism paramecium. Such cells swim about gracefully, avoid obstacles and predators, find food and engage in sex with partner paramecia. They can also learn; if placed in capillary tubes they escape, and when placed back in the capillary tubes escape more quickly. As single cells with no synaptic connections, how do they do it? Pondering the seemingly intelligent activities of such single cell organisms, famed neuroscientist C.S. Sherrington (1957) conjectured: “of nerve there is no trace, but the cytoskeleton might serve”. If the cytoskeleton is the nervous system of protozoa, what might it do for neurons?

IV. The neuronal cytoskeleton

a. Microtubules and networks inside neurons

Shape, structure, growth and function of neurons are determined by their cytoskeleton, internal scaffoldings of filamentous protein polymers which include microtubules, actin and intermediate filaments. Rigid microtubules (MTs) interconnected by MT-associated proteins (MAPs) and immersed in actin form a self-supporting, dynamic tensegrity network which shapes all eukaryotic cells including highly asymmetrical neurons. The cytoskeleton also includes MT-based organelles called centrioles which organize mitosis, membrane-bound MT-based cilia, and proteins which link MTs with membranes. Disruption of intra-neuronal cytoskeletal structures impairs cognition, such as tangling of the tau MAP linking MTs in Alzheimer’s disease (Matsuyama and Jarvik, 1989, Iqbal and Grundke-Iqbal 2004).

Actin is the main component of dendritic spines and also exists throughout the rest of the neuronal interior in various forms depending on actin-binding proteins, calcium etc. When actin polymerizes into a dense meshwork, the cell interior converts from an aqueous solution (sol state) to a quasi-solid, gelatinous (gel) state. In the gel state, actin, MTs and other cytoskeletal structures form a negatively-charged matrix on which polar cell water molecules are bound and ordered (Pollack 2001). Glutamate binding to NMDA and AMPA receptors triggers gel states in actin spines (Fischer et al 2000).

Neuronal MTs self-assemble, and with cooperation of actin enable growth of axons and dendrites. Motor proteins transport materials along MTs to maintain and regulate synapses. The direction and guidance of motor proteins and synaptic components (e.g. from cell body through branching dendrites) depends on conformational states of MT subunits (Krebs et al 2004). Thus MTs are not merely passive tracks but appear to actively guide transport. Among neuronal cytoskeletal components, MTs are the most stable and appear best suited for information processing Wherever cellular organization and intelligence are required, MTs are present and involved.

MTs are cylindrical polymers 25 nanometers (nm = 10-9 meter) in diameter, comprised of 13 longitudinal protofilaments which are each chains of the protein tubulin (Figure 8). Each tubulin is a peanut-shaped dimer (8 nm by 4 nm by 5 nm) which consists of two slightly different monomers known as alpha and beta tubulin, (each 4 nm by 4 nm by 5 nm, weighing 55,000 daltons). Tubulin subunits within MTs are arranged in a hexagonal lattice which is slightly twisted, resulting in differing neighbor relationships among each subunit and its six nearest neighbors (Figure 9). Thus pathways along contiguous tubulins form helical pathways which repeat every 3, 5 and 8 rows (the Fibonacci series). Alpha tubulin monomers are more negatively charged than beta monomers, so each tubulin (and each MT as a whole) is a ferroelectric dipole with positive (beta monomer) and negative (alpha monomer) ends.[xxiii]

In non-neuronal cells and in neuronal axons, MTs are continuous and aligned radially like spokes of a wheel emanating from the cell center. MT negative (alpha) ends originate in the central cell hub (near the centrioles, or MT-organizing-center adjacent to the cell nucleus) and their positive (beta) ends extend outward to the cell perimeter. This is the case in axons, where the negative ends of continuous MTs originate in the axon hillock, and positive ends reach the pre-synaptic region.

Figure 8. Microtubule (left) is a cylindrical polymer of subunit proteins known as tubulin arranged in a skewed hexagonal lattice. Each tubulin can exist in two or more conformational states, e.g. open (black) or closed (white). Right: Each tubulin state is governed by quantum mechanical London forces—collective positions of hundreds of electrons (represented here as two electrons) in nonpolar hydrophobic regons within the protein. Because of governance by quantum forces, it is proposed that tubulins can exist in quantum superposition of both conformations (black and white=gray). The actual displacement in the superposition separation need only be the diameter of a carbon atom nucleus, but is illsutrated here as roughly 10% of the protein volume.

However dendritic cytoskeleton is unique. Unlike axons and any other cells, MTs in dendrites are short, interrupted and mixed polarity. They form networks interconnected by MAPs (especially dendrite-specific MAP2) of roughly equal mixtures of polarity. There is no obvious reason why this is so—from a structural standpoint uninterrupted MTs would be preferable, as in axons. Networks of mixed polarity MTs connected may be optimal for information processing.  

Intra-dendritic MT-MAP networks are coupled to dendritic synaptic membrane and receptors (including dendritic spines) by mechanisms including calcium and sodium flux, actin and metabotropic inputs including second messenger signaling e.g. dephosphorylation of MAP2 (Halpain and Greengard 1990). Alterations in dendritic MT-MAP networks are correlated with locations, densities and sensitivities of receptors (e.g. Woolf et al 1999). Synaptic plasticity, learning and memory depend on dendritic MT-MAP networks.

Since Sherrington’s observation in 1957, the idea that the cytoskeleton—MTs in particular—may act as a cellular nervous system has occurred to many scientists. Vassilev et al (1985) reported that tubulin chains transmit signals between membranes, and Maniotis et al (1997a, 1997b) demonstrated that MTs convey information from membrane to nucleus. But MTs could be more than wires. The MT lattice is well designed to represent and process information, with the states of individual tubulins playing the role of bits in computers. Conformational states of proteins in general (e.g. ion channels opening/closing, receptor binding of neurotransmitter etc.) are the currency of real-time activities in living cells. Numerous factors influence a protein’s conformation at any one time, so individual protein conformation may be considered the essential input-output function in biology. 

b. Microtubule automata

The peanut-shaped tubulin dimer switches between two conformations in which the alpha monomer flexes 30 degrees from vertical alignment with the beta monomer. These are referred to as open and closed states (Figure 8, Melki et al 1989, Hoenger and Milligan 1997, Ravelli et al 2004).[xxiv]

Atema (1973) proposed that tubulin conformational changes propagated as signals along MTs in cilia. Hameroff and Watt (1982) suggested that the MT lattice acted as a two-dimensional computer-like switching matrix with tubulin states influenced by neighbor tubulins, and input/output occurring via MAPs.[xxv] MT information processing was also viewed in the context of cellular automata (Smith et al 1984, Rasmussen et al 1990).

Cellular automata are self-organizing information systems based on lattices of fundamental units (cells) whose states interact with neighbor cells at discrete time steps. In a two dimensional checkerboard lattice, each cell has eight neighbors (corner neighbors included) and exists in two (or more) possible states. Neighbor interaction rules determine each cell’s state at the next time step.

A well-known example is the Game of Life in which two possible states of each cell whimsically represent either alive or dead on a checkerboard lattice (Gardner 1970). There are three neighbor rules:

• ·        If the number of live neighbors is exactly two, the cell maintains status quo into the next generation. Thus if the cell is alive, it stays alive, if it is dead, it stays dead.

• ·        If the number of live neighbors is exactly three, the cell will be alive in the next generation regardless of the cell's current state. A dead cell is “born”, a live cell lives on.

• ·        If the number of live neighbors is 0, 1, or 4-8, the cell will be dead in the next generation due to not enough support (0 or 1) or overcrowding  (4-8).

The generations are synchronized by a universal clock mechanism. Starting from random initial patterns, complex behaviors emerge, for example chaotic dynamics (Wolfram 1984, Langton 1990). However common types of patterns generally appear: stable objects, oscillators/blinkers and gliders which move through the grid. Streams of gliders can perform all logic and memory functions on which computers are based. The Game of Life and cellular automata in general are universal computers.

MTs were modeled as automata in which tubulin conformational states (open, closed) interacted with neighbor tubulin states by dipole interactions. Dipole strengths in open and closed conformations were used to generate interaction rules. Thus the dipole-coupled conformation for each tubulin was determined at each generation by the sum of the dipoles of its six surrounding neighbors.[xxvi] Because of the skewed hexagonal geometry, contributions from each of the six neighbors differed (Figure 9). The generations, or time steps were assumed to be nanoseconds, following Fröhlich’s suggestion of coherent excitations.

Figure 9. The lattice of tubulins in microtubules. Left: The lattice showing the tubulin dimers as (negatively charged) alpha monomers and (positively charged) beta monomers. Middle: A tubulin neighborhood is defined by identifying the central tubulin C and its 6 surrounding neighbors by compass points: N (north), NE (northeast), SE (southeast), S (south), SW (southwest), NW (northwest). Right: The spacings (in nanometers) and definition of angle ?. y is the vertical distance between (the same points on) any two neighboring dimers and r the absolute distance. While y varies, the horizontal distance is always 5 nanometers. Curvature around the cylinder is ignored and the dipole force between dimers related to y/r3. From Rasmussen et al (1990).  

Figure 10. Cellular automata. Top two rows: Two different sequences of gliders moving in the Game of Life. In the first row the glider moves downward; in the second row the glider moves upward.   Bottom two rows: Two different sequences of gliders moving and patterns evolving in microtubule automata. In third row, gliders move downward through the microtubule; in the fourth row, patterns move both upward (black column, 4th protofilament) and downward (white column, 2nd protofilament).

Herbert Fröhlich (1968, 1970, 1975) proposed that a set of dipoles constrained in a common geometry and electric field would oscillate in phase, coherently like a laser[xxvii] if biochemical energy were supplied. Membrane proteins and tubulins in MTs are predicted to oscillate in the range of 10-9 to 10-11 seconds.[xxviii]

Simulations of MT automata showed stable patterns, blinkers and propagating gliders (velocity 8 to 800 m/sec,[xxix] Figure

10). Two MT automata interconnected by MAPs exhibited recognition and learning (Figure 11; Rasmussen et al 1990).

MT automata potentially increase cellular and brain-wide information processing enormously. Neurons each contain at least 10^7 tubulins (Yu and Baas 1994); switching in nanoseconds (109/sec) predicts roughly 1016 operations per second per neuron.[xxx] But enhanced information processing per se fails to answer fundamental questions about consciousness. A clue lies in the mechanism of switching in proteins.

Figure 11. Interior schematic of dendrite showing unique mixed polarity networks of microtubule automata interconnected by microtubule-associated proteins (MAPs). Inputs to microtubule automata (orchestration) from e.g. glutamate activation of dendritic spine receptors are conveyed by sodium and calcium ion flux along actin filaments. MAPs convey information between MTs to form an automaton network. Output/results of MT automaton network processing can trigger axonal spikes, regulate synapses and hardwire memory.

c. Protein conformational dynamics—Nature’s bits and qubits

Proteins are the engines of life, dynamically changing conformational shape at multiple scales (Karplus and McCammon 1983). Functional changes occur in 10-6 sec to 10-11 sec transitions. Proteins have large energies with thousands of kiloJoules per mole (kJ mol-1) but are only marginally stable against denaturation by ~40 (kJ mol-1). Consequently protein conformation is a "delicate balance among powerful countervailing forces" (Voet and Voet 1995).

Individual proteins are linear chains of amino acids which fold into three-dimensional conformations.[xxxi] The driving force in folding is joining together of uncharged non-polar amino acid groups, repelled by solvent water. These hydrophobic groups attract each other by van der Waals forces, avoiding water and burying themselves within protein interiors forming (in some proteins) hydrophobic pockets.[xxxii] Volumes of pockets (~0.4 cubic nanometers) are 1/30 to 1/250 the volume of single proteins. Though tiny, hydrophobic pockets are critically important in the determination of protein conformation both in folding and regulation of conformational dynamics. Hydrophobic pockets may act as the brain of a protein.

Non-polar (but polarizable) amino acid side groups within hydrophobic pockets interact by van der Waals London forces. Electrically neutral atoms and non-polar molecules can have instantaneous dipoles in their electron cloud distribution. Electrons in clouds from neighboring non-polar amino acid side groups repel each other, inducing mutual fluctuating dipoles which then couple to each other like oscillating magnets. As high energy forces cancel out, weak but numerous (thousands per protein) London forces govern protein conformation (Figure 8).[xxxiii]

Due to inherent uncertainty in electron localization, London forces are quantum mechanical effects. Thus proteins governed by London forces in hydrophobic pockets are quantum levers, amplifying quantum forces to govern conformational changes and physical effects. Prevention of quantum leverage accounts for the action of anesthetic gases.

d. Anesthesia

Millions of people every year undergo general anesthesia for surgery with complete and  reversible loss of consciousness. At a critical concentration of anesthetic drug, consciousness is erased while many nonconscious functions of brain and other organs continue (e.g. EEG, evoked potentials, control of breathing). How does this happen? 

The situation seems confusing, with many different types of drugs acting on many different types of proteins in the brain (e.g. receptors for various excitatory and inhibitory neurotransmitters, channels, enzymes, connexin in gap junctions (Masaki et al 2004, He & Burt 2000), actin, tubulin in microtubules). Purely inhalational anesthetic gases which travel through the lungs and blood to the brain constitute a variety of types of molecules: halogenated hydrocarbons, ethers, the inert element xenon, nitrous oxide etc. However there is one important unifying feature.

All anesthetic gas molecules are non-polar, and thus poorly soluble in water/blood, but highly soluble in a particular lipid-like, hydrophobic environment akin to olive oil. The potency of anesthetic gases in erasing consciousness correlates perfectly with solubility in such an environment. The brain has a large lipid-like (olive oil-like) domain, both in lipid regions of neural membranes and hydrophobic pockets within certain proteins. Anesthetics were originally thought to act in lipid regions of membranes, but protein hydrophobic pockets were determined to be their primary sites of action (Franks and Lieb 1982). Anesthetic gases bind to non-polar amino acid groups in the pockets (e.g. the benzene-like ring in phenylalanine, and the indole ring in tryptophan) by van der Waals London forces, the same quantum forces which form the pockets and govern conformational dynamics.

Why do weak quantum forces have such profound and selective effects? Anesthetic gas molecules form their own London force interactions with non-polar amino acid groups, preventing or altering normally occurring London forces necessary for protein conformational dynamics and consciousness. Anesthetic gases prevent quantum leverage.

Most protein conformational changes are unaffected by general anesthetics—muscle contractility, enzyme function and most brain activities (as evidenced by EEG and evoked potentials) continue during anesthesia. Axonal action potentials are also relatively unaffected by general anesthetics. Proteins which are affected include post-synaptic receptors (acetylcholine, serotonin, GABA and glycine), tubulin (Allison and Nunn 1968) and actin, which disassembles in dendritic spines when exposed to anesthetics (Kaech et al 1999).

Anesthetics act (and consciousness occurs) not in any one brain region, or in any one type of neuron or particular protein. Rather, anesthesia and consciousness occur in hydrophobic pockets of a class of proteins in dendrites throughout the brain (Hameroff 1998c). In these pockets, quantum London forces govern protein function responsible for consciousness. Does that imply that consciousness is a quantum process?

V. Quantum information processing

a. Quantum mechanics

Reality is described by quantum physical laws that reduce to classical rules (e.g. Newton’s laws of motion) at certain large scale limits. According to quantum physical laws:  

• Objects/particles may exist in two or more places or states simultaneously—more like waves than particles and governed by a quantum wavefunction. This property of multiple coexisting possibilities is known as quantum superposition.

• Multiple objects/particles can be unified, acting as a single coherent object governed by one wavefunction. If a component is perturbed, others feel it and react. This is called nonlocality and is the main difference between classical and quantum physics.  If the objects remain together, nonlocality is known as Bose-Einstein condensation.
• If unified objects are spatially separated they remain unified. This nonlocality is also known as quantum entanglement.

Why don’t we see quantum superpositions in our world? How are quantum particles connected over distance?

Experiments show that quantum superpositions persist until they are measured, observed  or interact with the classical environment (decohere). If such interactions occur, quantum superpositions reduce, collapse or decohere to particular classical states, with the particular choice of states apparently random. What actually constitutes the act of measurement/observation is unclear, as is the fate of isolated, unmeasured quantum superpositions. Interpretations of quantum mechanics address this issue:

• The Copenhagen interpretation (measurement or conscious observation collapses the wave function)[xxxiv] puts consciousness outside physics and doesn't account for fundamental reality.
• The multiple worlds view suggests each superposition is amplified, leading to a new universe. There is no collapse, but an infinity of realities (and conscious minds) is required.
• David Bohm’s interpretation avoids reduction/collapse but requires another layer of reality. Matter exists as objects guided by complex waves of possibility.
• Henry Stapp views the universe as a single quantum wave function. Reduction of part of it within the brain is a conscious moment (akin to Whitehead’s “occasion of experience” – Whitehead 1929, 1933). Reduction/collapse is consciousness, but its cause and distinction between universal wave function and that within the brain are unclear.
• In decoherence theory any interaction (loss of isolation) of a quantum superposition with a classical system (e.g. through heat, direct interaction or information exchange) erodes the quantum system. But 1) the fate of isolated superpositions is not addressed, 2) no quantum system is ever truly isolated, 3) decoherence doesn’t actually disrupt superposition, just bury it in noise, 4) some quantum processes are enhanced by heat and/or noise.
• An objective threshold for reduction (objective reduction, OR) exists due to e.g. the number of superpositioned particles (GRW theory – Ghirardi et al 1990) or quantum gravity as in the OR proposals of Károlyházy et al (1986), Di¢si (1989) and Roger Penrose (1989).

How can objects actually be in multiple locations or states simultaneously? Penrose (1989, 1994) takes superposition as an actual separation in underlying reality at its most basic level (fundamental spacetime geometry at the Planck scale of 10-33 cm).[xxxv] This is akin to the multiple worlds view (superpositions are amplified to form a separate  universe), however according to Penrose the separations are unstable and (instead of branching off completely) spontaneously reduce (self-collapse) due to an objective feature of spacetime geometry.[xxxvi] Accordingly, the larger the superposition, the more rapidly it reduces. For example an isolated one kilogram object in superposition would meet OR quickly, in only 10-37 seconds. An isolated superpositioned electron would undergo OR only after 10 million years. Penrose OR is currently being tested experimentally (Marshall et al 2003).

In The emperor's new mind Penrose (1989) suggested that choices resulting from this OR were not random (as are those from measurement and decoherence) but influenced by Platonic information embedded at the Planck scale, the fundamental level of the universe. Moreover this particular type of non-random, non-algorithmic (non-computable) selection is characteristic of conscious choices, differing in a basic way from the output of classical computers. Therefore Penrose proposed that OR-mediated quantum computation must be occurring in the brain. Quantum computation (see next Section) relies on both superposition and entanglement.

Entanglement is even stranger than superposition. Quantum theory predicted that  complementary quantum particles (e.g. electrons in coupled spin-up and spin-down pairs) would remain entangled even when separated. Einstein, Podolsky and Rosen (1935) described a thought experiment to disprove this notion (Figure 3). An entangled complementary pair of superpositioned electrons (EPR pairs) would be separated and sent in different directions along two different wires, each electron remaining in superposition. When one electron was measured at its destination and, say, spin-up was observed, its entangled twin miles away would correspondingly reduce instantaneously to spin-down which would be confirmed by measurement. This would require a faster-than-light signal which Einstein’s special relativity had precluded. Nonetheless since the early 1980s (Aspect et al 1982, Tittel et al 1998) this type of experiment has been performed through wires, fiber optic cables and via microwave beams through atmosphere. Entanglement has been repeatedly confirmed. The mechanism of instantaneous communication remains unknown, seeming to violate special relativity.

To explain entanglement, Penrose (2004, c.f. Bennett & Wiesman 1992) suggested backward time referral of quantum information, i.e. from the measurement back in time to the unified complementary pair, then forward in time to the opposite twin (Figure 3). In the quantum world, time is symmetric (bidirectional), or the flow of time doesn’t exist.

Although poorly understood, entanglement and superposition are used in quantum computing and related technologies.     

b.      Quantum computation

Proposed in the 1980s (independently) by Benioff (1982), Deutsch (1985) and Feynman (1986), quantum computers (and quantum cryptography and teleportation) are being developed in a variety of technological implementations.

The basic idea is this. Conventional computers represent digital information as binary bits of either 1 or 0. Quantum computers can represent quantum information as superpositions of both 1 AND 0 (quantum bits, or qubits). While in superposition (and isolated from environment) qubits interact with other qubits by nonlocal entanglement, allowing interactions to evolve[xxxvii] resulting in computation of enormous speed and near-infinite parallelism. After the interaction/computation is performed, qubits are reduced/collapsed to specific classical bit states by measurement, giving the output or solution.[xxxviii]

The major hurdle to quantum computing is the sensitivity of quantum superpositions to disruption by thermal vibration or any interaction with the environment—decoherence. Consequently quantum computing prototypes have been built to operate at extremely cold temperatures to avoid thermal noise, and in isolation from the environment.

In the mid 1990s quantum error correcting codes were developed which could detect and correct decoherence, preserving the quantum information (Steane 1998). Topological quantum error correction was developed in which the geometry of the quantum computer lattice was inherently resistant to decoherence. For example a quantum computer could utilize the Aharonov-Bohm effect in which alternate possible paths of a quantum particle are considered as a superposition of paths (Kitaev 1997). So lattice pathways (rather than individual components of those pathways) can be global qubits resistant to decoherence.

c. Quantum computing with Penrose OR

Technological qubits reduce/collapse by measurement, bringing in a component of randomness averaged out by redundancy. According to Penrose (1989) quantum computation which self-collapses by OR avoids randomness, instead providing a non-computable influence stemming from Platonic values embedded at the Planck scale. Such quantum computation would be algorithmic up to the instant of OR, with an added modification then occurring.

The Penrose argument for non-computability using Gödel’s theorem was harshly criticized but not refuted. For consciousness, OR also provides explanations for:  

• Transition from nonconscious (superpositioned quantum information) to classical information, with consciousness the transition itself.
• Binding via quantum coherence, condensation and/or entanglement
• Libet’s backward time referral and other temporal anomalies.
• The hard problem of conscious experience via Whitehead pan-protopsychism connected to fundamental spacetime geometry (Hameroff and Penrose 1996b).

Penrose initially had no clear candidate for biological qubits in the brain, suggesting only the possibility of superpositions of neurons both firing and not firing. Microtubules seemed ideal for the type of quantum computation Penrose was suggesting. 

Penrose implied that nonconscious processes capable of becoming conscious utilize quantum information. What do we know about nonconscious[xxxix] processes?

VI. The quantum subconscious

German psychologist Frederic Meyer in 1886 described subliminal consciousness, followed by William James’ transmarginal consciousness or fringe, a region of the mind just outside consciousness but accessible to it.

Sigmund Freud saw dreams as the “Royal road to the unconscious” whose bizarre character was due to censorship and disguise of thwarted drives. Freud’s ideas became downplayed, and dreams characterized as mental static (e.g. Hobson 1988, 2004). However recent brain imaging shows dream-associated REM sleep activity in regions associated with emotion and gratification (Solms 2000, 2004).

Chilean psychologist Ignacio Matte Blanco (1975, c.f. Rayner 1995) compared logic structure in dreams to the Aristotelian logic of waking consciousness in which, for example, the logic statement: 

If x, then y

does not imply the statement:

If y then x. 

This is obvious to our conscious minds. For example:

If the light turns green, then I go

Does not imply:

If I go, then the light will turn green.  

However from decades of dream analysis Matte Blanco determined two non-Aristotelian axioms of the logic of the unconscious: symmetry and generalization. In dreams:

If x then y

implies that also

If y then x.

In dreams, according to Matte Blanco:

If the light turns green, then I go  

implies that also:

If I go, then the light turns green.

Generalization means that any entity is a part of a whole, and when symmetry and generalization are combined, paradox occurs. For example:

If a hand is part of the body

then also:

The body is part of the hand. 

The seeming contradiction of any set being a subset of itself defines an infinite set, and is also holographic (and fractal). Any part of a whole also contains the whole within the part.[xl]

Symmetry also means that:

If event a happened after event b, 

then also:

Event b happened after event a.   

From this Matte Blanco concludes: “..the processes of the unconscious …are not ordered in time.”

Another implication of unconscious logic is that apparently negating propositions (e.g. p and not p) may be true, resulting in coincidence of contraries. For example (to use Matte Blanco’s example):

x is alive

and

x is dead

are both true (e.g. when time is removed. More generally, according to Matte Blanco, “the unconscious is unable to distinguish any two things from each other”.

The unconscious utilizes multiple coexisting possibilities, inseparability and timelessness, very much like quantum information. Matte Blanco summarized the unconscious as “where paradox reigns and opposites merge to sameness”, also an apt description of the quantum world.

VII. Quantum computation in microtubules—The Orch OR model

a.       Specifics of Orch OR

In a proposal for the mechanism of consciousness, Roger Penrose and I have suggested that microtubule (MT) quantum computations in neurons are orchestrated by synaptic inputs and MT-associated proteins (MAPs), and terminate (e.g. after 25 msec, 40 Hz) by Roger’s objective reduction OR mechanism. Hence the model is known as orchestrated objective reduction, Orch OR. Complete details may be found in Penrose and Hameroff (1995), Hameroff and Penrose (1996a & 1996b) and Hameroff (1998a). The key points are:

1)      Conformational states of tubulin protein subunits within dendritic MTs interact with neighbor tubulin states by dipole coupling such that MTs process information in a manner analogous to cellular automata which regulate neuronal activities (trigger axonal spikes, modify synaptic plasticity and hardwire memory by MT-MAP architecture etc.).

2)     Tubulin conformational states and dipoles are governed by quantum mechanical London forces within tubulin interiors (non-polar hydrophobic pockets) so that tubulins may exist as quantum superpositions of differing conformational states, thus acting as quantum levers and qubits.[xli]

3)      While in superposition, tubulin qubits communicate/compute by entanglement with other tubulin qubits in the same MT, other MTs in the same dendrite, and MTs in other gap junction-connected dendrites (i.e. within a hyper-neuron). Thus quantum computation occurs among MTs throughout macroscopic regions of brain via tunneling through gap junctions or other mechanisms.[xlii]

4)      Dendritic interiors alternate between two states determined by polymerization of actin protein: 1) In the liquid (solution: sol) state, actin is depolymerized and MTs communicate/process information classically (tubulin bits) with the external world. During this phase synaptic activities provide inputs via MAPs which orchestrate MT processing and (after reduction) MT (output) states regulate axonal firing and synaptic plasticity. 2) As actin polymerizes (e.g. triggered by glutamate binding to receptors on dendritic spines), dendritic cytoplasm enters a quasi-solid gelatinous (gel) state, MTs become isolated from environment and enter quantum superposition mode in which tubulins function as quantum bits or qubits (Figure 12). The two states alternate e.g.at 40 Hz.

5)      Quantum states of tubulin/MTs in gel phase are isolated/protected from environmental decoherence by biological mechanisms which include encasement by actin gelation, ordered water, Debye screening, coherent pumping and topological quantum error correction (Section VIIb).

6)      During quantum gel phase, MT tubulin qubits represent pre-conscious information as quantum information—superpositions of multiple possibilities, of which dream content is exemplary.

7)      Pre-conscious tubulin superpositions reach threshold for Penrose OR (e.g.after 25 msec) according to E=h/t in which E is the gravitational self-energy of the superpositioned mass (e.g. the number of tubulins in superposition),  h is Planck’s constant over 2π, and t is the time until OR. Larger superpositions (more intense experience) reach threshold faster. For t=25 msec (i.e. 40 Hz) E is roughly 1011 tubulins, requiring a hyper-neuron of minimally 104 neurons per conscious event (Hameroff and Penrose 1996a). The makeup of the hyper-neuron (and content of consciousness) evolves with subsequent events.

8)      Each 25 msec OR event chooses ~1011 tubulin bit states which proceed by MT automata to govern neurophysiological events, e.g. trigger axonal spikes, specify MAP binding sites/restructure dendritic architecture, regulate synapses and membrane functions. The quantum computation is algorithmic but at the instant of OR a non-computable influence (i.e. from Platonic values in fundamental spacetime geometry) occurs.

9)      Each OR event ties the process to fundamental spacetime geometry, enabling a Whiteheadian pan-protopsychist approach to the 'hard problem' of subjective experience. A sequence of such events gives rise to our familiar stream of consciousness.

Applications of Orch OR to aspects of consciousness and cognition will be considered in Section VIII.

b. Decoherence

Decoherence is the disruption of quantum superposition due to energy or information interaction with the classical environment. Consequently quantum technology is generally developed in ultra-cold isolation, and physicists are skeptical of quantum computing in the “warm, wet and noisy” brain.

However biological systems may delay decoherence in several ways (Davies 2004). One is to isolate the quantum system from environmental interactions by screening/shielding. Intra-protein hydrophobic pockets are screened from external van der Waals thermal interactions; MTs may also be shielded by counter-ion Debye plasma layers (due to charged C-termini tails on tubulin) and by water-ordering actin gels (Hameroff et al 2002). Biological systems may also exploit thermodynamic gradients to give extremely low effective temperatures (Matsuno 1999).

Another possibility concerns decoherence-free subspaces. Paradoxically, when a system couples strongly to its environment through certain degrees of freedom, it can effectively “freeze” other degrees of freedom (by a sort of quantum Zeno effect), enabling coherent superpositions and entanglement to persist (Nielson & Chuang 2001). Metabolic energy supplied to MT collective dynamics (e.g. Fröhlich coherence) can counter decoherence (in the same way that lasers avoid decoherence at room temperature). Finally, MT structure seems ideally suited for topological quantum error correction by the Aharonov-Bohm effect (Hameroff et al 2002).

Attempting to disprove a role for quantum states in consciousness, Max Tegmark (2000, c.f. Seife 2000) calculated MT decoherences times of 10^-13 sec, far too brief for neural activities. However Tegmark did not address Orch OR nor any previous proposal, but his own quantum MT model which he did indeed successfully disprove. Hagan et al (2002) recalculated MT decoherence times with Tegmark’s formula[xliii] but based on stipulations of the Orch OR model. For example Tegmark used superposition of solitons “separated from themselves” along a microtubule by a distance of 24 nanometers. In Orch OR, superposition separation distance is the diameter of a carbon atom nucleus, 6 orders of magnitude smaller. Since separation distance is in the denominator of the decoherence formula, this discrepancy alone extends the decoherence time 6 orders of magnitude to 10^-7 seconds. Additional discrepancies (charge versus dipole, correct dielectric constant) extend the calculated decoherence time to 10^-5 to 10^-4 sec. Shielding (counter-ions, actin gel) extends the time into physiological range of tens to hundreds of msec. Topological (Aharonov-Bohm) quantum error correction may extend MT decoherence time indefinitely.[xliv]

Is the brain truly “wet and noisy”? In gel state MTs are in a quasi-solid environment with ordered water. As for “noisy”, electrophysiological background fluctuations show ongoing “noise” to actually correlate over distances in the brain (Arieli et al 1996, Ferster 1996).

Experimental evidence shows that electron quantum spin transfer between quantum dots connected by organic benzene molecules is more efficient at room temperature than at absolute zero (Ouyang and Awschalom, 2003). The same structures are found in amino acids (phenylalanine, tyrosine, tryptophan) in hydrophobic pockets of proteins. Other experiments have shown quantum wave behavior of biological porphyrin molecules (Hackermüller et al., 2003), and still others that noise can enhance some quantum processes (Beige et al 2004). Evolution has had billions of years to solve the decoherence problem (Section IXf). 

Figure 12. Interior schematic of dendrites in quantum isolation phase. Actin has polymerized into gel form and MAPs detached, shielding and isolating MTs whose tubulins have evolved into quantum superposition.

c. Testability and falsifiability

In 1998 twenty testable predictions of Orch OR were published (Hameroff 1998a). Of these the following have been validated: signaling along MTs (Maniotis et al 1997a, 1997b), correlation of synaptic function/plasticity with cytoskeletal structure (Khuchua et al 2003, Woolf 1998, O'Connell et al 1997), actions of psychoactive drugs involve MTs (Andrieux et al 2002), and gap junctions mediate gamma synchrony/40 Hz (numerous references cited in Section IIIe). Others are currently being tested, and all are listed in the endnotes.[xlv] None have as yet been proven wrong. With the possible exception of the link to Planck scale geometry, all are imminently testable. Orch OR is falsifiable—it need only be shown that consciousness can occur without dendrites, gap junctions (or some other mechanism for brain-wide quantum coherence), microtubules or quantum computation and Orch OR is falsified

Figure 13. Conscious events. Top: Microtubule automata enter pre-conscious quantum superposition phase (gray tubulins) until threshold for OR is met after 25 msec (this would involve superposition of 1011 tubulins in tens of thousands of neurons interconnected by gap junctions). A conscious moment (NOW) occurs, new classical states of tubulins are chosen and a new sequence begins. Middle: Phase diagram of increasing superposition in gel phase which meets threshold after e.g. 25 msec. A conscious event (NOW) occurs, and the cycle repeats. Bottom: After each OR event, quantum information is sent backward in time to influence previous event. Classical information (memory) goes forward in time.

VIII.         Applications of Orch OR to consciousness and cognition

a.       Visual consciousness

Visual components (e.g. shape, color, motion) are processed in separate brain areas and at different times but integrated into unified visual gestalts. How does this occur? And how do 40 Hz excitations relate to longer periods associated with the visual gestalt (e.g. 250 to 700 msec)?

Thalamic inputs to V1 are fed-forward to areas V2, V3, V4 and LO for shape recognition, then to V8 and V4v for color, and to V5, V3A and V7 for motion, then back to V1 and pre-frontal cortex. In Woolf and Hameroff (2001) we suggested that these component steps each correspond with 40 Hz excitations, and microconsciousness as proposed by Zeki (2003). To unify components in a visual gestalt after 250 msec, a cumulative snowball effect—a crescendo of crescendos—occurs (corresponding with the growth of a hyper-neuron, Figure 14). Commenting on this proposal Gray (2004) points out that we are conscious only of the visual gestalt, not incremental components. This suggests that backward referral enables each OR event to refer quantum information backward in time (the duration proportional to E). Thus quantum information/qualia of visual components are referred back to the initial V1 potential, resulting in an integrated visual gestalt early in the integration process.

Consequently tennis and baseball players consciously see and recognize the ball’s shape, color and motion early enough to respond successfully. In the color phi phenomenon the brain fills in the gap by backwards referral from the subsequent location. Thus unlike retrospective construction, conscious sensation actually occurs in transit between the two locations.[xlvi]

Figure 14. Visual gestalts. Left: A crescendo sequence of ~25 msec/40 Hz quantum computations/conscious events of components of conscious vision culminating in an integrated visual gestalt after e.g. 250 to 700 msec (modified from Woolf and Hameroff 2001). The intensity (y axis) is related to the amount of superposition represented by E=h/t. Thus the slope/intensity for each event is inversely proportional to time to OR. Right: Modified version in which components are referred backward in time as nonconscious quantum information. The duration backward in time is related to slope/intensity of each component event. Thus an integrated visual gestalt occurs early in visual processing.   

b. Volition and free will

Volition and free will raise two major issues. One is time, in which we apparently act prior to processing the relevant inputs to which we respond. Backward time referral of unconscious quantum information can solve this problem.

The other issue is determinism. If brain processes (including nonconscious processes) and events in our environment are algorithmic—even if highly nonlinear/chaotic—then our actions are deterministic products of genetic influences and experience. Wegner (2002) concludes that free will is the (illusory) conscious experience of acting deterministically. The non-computable aspect of Penrose OR can help. 

Suppose I am playing tennis about to return my opponent’s ground stroke. As I begin to get my racket in position, I consider hitting a) to his forehand, b) to his backhand, c) a drop shot. A quantum superposition of these three possibilities (manifest as tubulin qubits) in a pre-motor cortical hyper-neuron evolves and reaches threshold for OR, at which instant one set of tubulin states corresponding with one action (e.g. hit to his forehand) is chosen resulting in the appropriate set of axonal spikes to execute the choice.

Could such actions be completely algorithmic and classical? Yes, but in addition to the beneficial time effect, the non-computable influence in Penrose OR can provide intuition, tipping the balance to the appropriate choice.[xlvii] Sometimes (it seems to me at least) we do things and we’re not quite sure why we do them.

This is not free will in the sense of complete agency because the non-computable influence is ultimately deterministic.[xlviii] What we experience as free will is algorithmic processes influenced by non-computable factors. This differs from Wegner’s (2002) view in that 1) our actions are not completely algorithmic, and 2) because of backward time referral, decisions are made consciously, concomitantly with the experience of the choice and action, and 3) consciousness is not epiphenomenal.  

c. Quantum associative memory

Evidence suggests memory is hard-wired in dendritic cytoskeletal structure (Khuchua et al 2003, Woolf 1998, O'Connell et al 1997). Woolf and Hameroff (2001) suggested perception of a stimulus precipitates conscious awareness of associated memory via EPR-like OR of entangled (associated) information. This implies that disparate contents of unified consciousness remain entangled in memory (Hameroff 2004).  

d. The hard problem of conscious experience

How the brain produces phenomenal experience composed of qualia—the smell of a rose, the felt qualities of emotions, and the experience of a stream of conscious thought—is the ‘hard problem’ (Chalmers 1996).

Broadly speaking, there are two scientific approaches: 1) emergence (experience arises as a novel property from complex interactions among simple components in hierarchical, recursive systems), and 2) some form of panpsychism, pan-protopsychism, or pan-experientialism (essential features or precursors of conscious experience are fundamental components of reality, accessed and organized by brain processes).   

Emergence derives from the mathematics of nonlinear dynamics, e.g. describing weather patterns, candle flames and self-organizing computer programs. Is consciousness an emergent property of interactions among neurons (or among tubulin proteins in microtubul