Summary & Insights
0:00:03 – Hi, and welcome to the A16Z podcast.
0:00:04 I’m Hannah.
0:00:07 In this episode, we talk all about the biology of pain
0:00:09 with Clifford Wolfe, professor of neurobiology
0:00:10 at Harvard Medical School,
0:00:13 who has dedicated his career to the subject.
0:00:15 The conversation covers everything you ever wondered
0:00:17 about the nature of pain, from why we even feel it,
0:00:20 and what the biological purpose of pain is,
0:00:22 to whether or not all pain is the same,
0:00:25 or if some kinds of pain function differently,
0:00:27 feel different, have different purposes,
0:00:30 or even if some people feel pain differently from others.
0:00:33 Wolfe describes the four different broad types
0:00:36 of pain we experience, what the purpose of each one is,
0:00:38 what it means now that we can phenotype
0:00:41 different kinds of pain and begin to understand them
0:00:43 as distinct and different from each other,
0:00:46 and how technology is enabling this new, deeper,
0:00:48 and much more complex understanding
0:00:50 of the nature of pain.
0:00:52 We also talk about what the biological link
0:00:54 between pain and addiction really is,
0:00:56 plus the one thing you might do
0:01:00 that we know makes your experience of pain worse.
0:01:01 You’ve been working on the subject of pain
0:01:03 for your entire career.
0:01:04 What drew you to that subject to begin with?
0:01:07 What was it that first sparked that initial interest for you
0:01:09 and where you began investigating?
0:01:11 – When I was a medical student,
0:01:15 I was on the surgical ward and at that time,
0:01:18 there was minimal treatment for postoperative pain,
0:01:21 so I came into the ward and there were a group of patients
0:01:24 who had major surgery,
0:01:28 and all of them were in severe discomfort,
0:01:31 and I said to the attending surgeon,
0:01:31 “What are you doing?
0:01:33 “Why aren’t you treating them with pain?”
0:01:35 He looked at me as if I was crazy,
0:01:36 and he said, “They’ve just had surgery.
0:01:38 “What do you expect? They’re in pain.”
0:01:41 – Even the idea of treating the pain was not–
0:01:42 – Absolutely, I was just thought,
0:01:43 “You have surgery, you have pain,”
0:01:45 and then you put up with it until the pain goes,
0:01:48 and I just thought to myself,
0:01:49 “That doesn’t sound right.”
0:01:51 At that time, there was very little understanding
0:01:53 of the mechanisms of pain,
0:01:56 and indeed, very few treatment options.
0:02:00 Now, as we have much greater understanding,
0:02:02 but unfortunately, the treatment options
0:02:03 haven’t really expanded,
0:02:08 and they are also accompanied by undesired effects,
0:02:10 particularly in the case of the opioids.
0:02:12 – So let’s talk a little bit about
0:02:14 what the understanding looked like then,
0:02:16 and what it’s beginning to look like now.
0:02:17 So what was our understanding
0:02:19 of the mechanism of pain back then
0:02:20 when we just sort of assumed,
0:02:22 “Okay, you have surgery, you’re in pain?”
0:02:25 – At that time, I just thought that all pain is similar,
0:02:27 that you either have no pain or you have pain,
0:02:31 and if you have pain, it’s an unpleasant sensation.
0:02:34 The closest analogy would be like flicking a switch
0:02:38 that switched on the pain sensation in your brain.
0:02:39 – So we thought it was that simple.
0:02:41 – We thought there was a peripheral trigger
0:02:43 which could be mild, like a pinprick,
0:02:45 or touching something too hard or too cold,
0:02:50 or larger in the sense of major trauma or post-surgical.
0:02:53 There was something that was activating that,
0:02:57 the pain switch, and that all the triggers of pain
0:02:59 acted on a single system in a single way,
0:03:02 and therefore pain really was not present,
0:03:04 or mild, moderate, or severe.
0:03:08 And that was the full range of the understanding of pain.
0:03:09 – Describe a little bit what the thought was
0:03:10 around that system.
0:03:12 Like how would that system be activated,
0:03:15 or what was our understanding of that system?
0:03:17 – Well, it certainly was appreciated
0:03:19 that pain has two facets.
0:03:24 One, it is an essential early warning device.
0:03:27 It tells us of danger in the environment.
0:03:30 And without it, there’s a high risk
0:03:31 that we may injure ourselves.
0:03:34 So you need this alarm system, if you like.
0:03:35 – And in the case of surgery,
0:03:38 that alarm system is going because you’re vulnerable,
0:03:39 you’re exposed, you’re healing.
0:03:42 – The alarm system is activated by the surgeon’s damage
0:03:45 that is created in the patient.
0:03:48 And then the alarm system continues
0:03:51 until that tissue injury is repaired.
0:03:53 But what is happening in a patient
0:03:55 who has persistent pain for years and years and years?
0:03:59 Why is that pain alarm activated continuously?
0:04:01 And there really was no explanation at the time.
0:04:04 – So that’s where you began your research?
0:04:05 – Exactly.
0:04:07 And at that time, if you had mild pain,
0:04:09 then you took a non-steroidal anti-inflammatory drug
0:04:12 like Tylenol or Advil.
0:04:14 If you had more severe pain,
0:04:17 you added a weak opioid to your NSAID.
0:04:19 And if you had severe pain,
0:04:22 then that’s the point at which you’d give a stronger opioid.
0:04:26 And the World Health Organization had an analgesic ladder
0:04:29 that it literally was that mild, moderate, severe pain
0:04:31 increasing the strength of the analgesic.
0:04:33 And again, in keeping with this notion
0:04:35 that pain was a unitary system activated
0:04:38 by different triggers of different intensity.
0:04:39 – It’s so simple.
0:04:42 It makes me think of the signs in the nurse’s doctor’s office
0:04:43 with the smiling faces.
0:04:45 Like even that is a whole order of magnitude
0:04:47 above in terms of understanding.
0:04:51 – Unfortunately, the reality is those smiling faces
0:04:55 would not be smiling if they saw how complicated pain was.
0:04:59 The notion that there’s a simple switch is just incorrect.
0:05:02 The notion that pain can be defined by its severity
0:05:05 as mild, moderate, or severe is incorrect.
0:05:07 And the notion that the treatment should be based on that
0:05:09 is also incorrect.
0:05:11 – What is it that has shifted in our understanding
0:05:14 of the biology of how much more complex it is?
0:05:17 – Pain as an early warning device,
0:05:20 kind of radar for danger, if you like,
0:05:23 is completely different from the pain that occurs
0:05:26 in the setting of a patient who has persistent pain.
0:05:28 – The quality of the pain?
0:05:30 – No, that was the confusing element.
0:05:32 We use the same word pain to describe both
0:05:35 and therefore physicians and patients felt
0:05:36 this is the same phenomenon.
0:05:39 But in fact, the pain that we experience
0:05:43 does not reflect what is initiating and sustaining the pain.
0:05:47 In the case of pain as a protective mechanism,
0:05:49 you need just an intense stimulus,
0:05:52 one of sufficient intensity
0:05:55 that potentially can damage our tissues.
0:05:57 And we have this very elaborate system
0:06:00 that has been designed to detect that danger.
0:06:03 So we have specialized sensory neurons
0:06:06 that are activated only by intense stimuli.
0:06:08 And when they reach that threshold,
0:06:11 which is just below the point at which damage can occur,
0:06:15 we feel a pin prick or the point at which something
0:06:19 that is too warm and it has the risk of burning us.
0:06:21 – And is that threshold the same for everyone
0:06:22 or different for everyone?
0:06:26 – It’s surprisingly similar that if you measure it
0:06:31 and in a situation where you remove all the emotional
0:06:35 aspects of pain, the threshold at which someone feels
0:06:38 something has been warm and then hotter.
0:06:40 And then when it switches to actually painful,
0:06:45 is it 42 degrees with a surprisingly small variation?
0:06:46 – That is surprising.
0:06:50 So the idea of being more pain tolerant
0:06:54 because of your accumulated experiences or sensitivities,
0:06:55 is that not…
0:06:57 – That is true, but that is not for that kind of pain.
0:07:01 If you’re drinking your coffee and it’s too hot
0:07:02 and it burns your tongue,
0:07:04 you will stop drinking it immediately.
0:07:05 You weren’t even aware of it.
0:07:08 We’ve all learned to take a little sip
0:07:11 and maybe not even put the coffee in our mouth.
0:07:13 So we’ve been trained by our experience.
0:07:15 We try and avoid that pain.
0:07:17 For that protective pain to work,
0:07:19 the pain has to be unpleasant.
0:07:21 And in order for it to work,
0:07:22 you can’t say, I’ll deal with it later.
0:07:25 You’ve got to deal with it immediately.
0:07:28 So it demands our attention and it’s intense
0:07:30 and it’s unpleasant.
0:07:32 And in that sense, it is that analogy of a switch.
0:07:35 You have switched on something that focus your attention.
0:07:36 I have to do something.
0:07:39 I have to avoid this situation immediately.
0:07:41 We’re talking a matter of seconds.
0:07:42 – It’s actually incredibly effective
0:07:45 when you think about it like that, a very effective tool.
0:07:46 – The protective mechanism of pain
0:07:49 requires an intense stimulus to activate it,
0:07:50 a noxious stimulus.
0:07:53 And we actually call this nociceptive pain,
0:07:56 which is the pain that is initiated by a noxious stimuli.
0:07:58 Now, the big difference
0:08:00 coming back to our post-surgical pain,
0:08:03 we now no longer dealing with protection from injury
0:08:04 because the injuries occurred.
0:08:06 It could either occur dramatically
0:08:08 if you’re involved in a motor vehicle accident
0:08:10 or it could be because you have surgery.
0:08:12 In both cases, there’s trauma to your tissue.
0:08:15 So you no longer need a protective system
0:08:18 to prevent tissue damage because you have tissue damage.
0:08:20 So what happens there?
0:08:23 And instead of protecting you from tissue damage,
0:08:26 evolution has developed a system
0:08:29 that helps the injury to heal.
0:08:31 And how does it do that?
0:08:34 It makes the injured part hypersensitive.
0:08:37 If you think about an injury or an inflammation,
0:08:40 just touching that wounded site will produce pain.
0:08:43 So we’ve now got a complete shift.
0:08:46 The pain is now activated by a stimulus
0:08:48 that would normally be innocuous,
0:08:51 a sensation that would normally just be light touch,
0:08:53 or in the case of a bad sunburn,
0:08:57 a shower that would normally be pleasantly warm
0:08:58 is now burning hot,
0:09:01 even though the temperature is not in the hot range.
0:09:03 – So it’s not about the stimulus,
0:09:04 it’s about the condition.
0:09:05 – Absolutely.
0:09:06 The nervous system has changed.
0:09:08 It’s become hypersensitive,
0:09:11 such that normally innocuous stimuli
0:09:14 are now activating the pain system,
0:09:17 and we have a tenderness, soreness,
0:09:19 a heightened sense of pain.
0:09:21 This is an adaptive thing.
0:09:25 For example, after major abdominal surgery,
0:09:27 if there were a way to completely eliminate the pain
0:09:30 and you say, “Fine, I’ll go for a jog,”
0:09:31 that would not be a good idea
0:09:33 because you’ve got this wounded tissue,
0:09:37 it needs time for the wound to heal, to repair itself.
0:09:40 And the same with if you’ve got arthritis,
0:09:42 where there’s damage to the joint,
0:09:43 if you excessively use the joint,
0:09:46 if you were pain-free, you would damage that joint.
0:09:49 – It’s still essentially protective in nature.
0:09:52 – It’s not protecting you from the external environment
0:09:54 and potential danger to your tissue,
0:09:59 it’s protecting the injured body part from further injury,
0:10:01 allowing repair to occur.
0:10:03 And so, finally, the pain that I thought
0:10:05 I first wanted to cure, post-operative pain
0:10:07 or post-traumatic pain,
0:10:09 so you do want to try and reduce it
0:10:10 to make people feel comfortable,
0:10:12 but at the same time, if you eliminated,
0:10:14 that would be a real problem.
0:10:17 But that then leaves our patients
0:10:19 who may not have any surgery,
0:10:21 they have no tissue damage,
0:10:24 and who have pain that just persists
0:10:26 for years and years and years.
0:10:28 One of the commonest causes of that
0:10:30 is damage to the nervous system itself,
0:10:32 and we call that neuropathic pain.
0:10:36 And we now appreciate this is no longer telling us
0:10:38 about the presence of noxious stimuli
0:10:40 or the presence of an injured part,
0:10:43 it is the malfunction of the nervous system itself,
0:10:45 it is a disease of the nervous system,
0:10:48 and it’s mechanistically quite different
0:10:50 from either the two other kinds of pain.
0:10:53 – So there’s three major categories of pain now.
0:10:54 – And there’s one fourth one.
0:10:55 – Oh, really?
0:10:58 – And this is a group of patients
0:11:02 who tend not exclusively, but tend to be women,
0:11:04 there’s no noxious stimuli,
0:11:06 there’s no tissue inflammation,
0:11:08 there’s no damage to the nervous system,
0:11:12 but they may have chronic widespread pain.
0:11:15 And for a long time, this was labeled as psychosomatic,
0:11:18 and there was almost a blame feature
0:11:21 to that label to the patients.
0:11:25 But now we realize they too have an abnormal nervous system.
0:11:27 The reason they’re feeling pain
0:11:30 is because it is if someone has switched up
0:11:34 the volume control in their central nervous system,
0:11:37 such that what would normally be a soft sound
0:11:40 is now very loud and for them,
0:11:43 a soft touch now produces pain.
0:11:46 So this is a pathological functioning of the nervous system
0:11:49 in the absence of damage to the nervous system.
0:11:51 – Is it congenital in those cases?
0:11:54 – Overall, for most clinical pain conditions,
0:11:59 there is actually a very strong heritable component,
0:12:03 something like 50% of the risk of developing pain
0:12:04 is heritable.
0:12:05 We know that from twin studies.
0:12:08 So if you look at identical twins,
0:12:12 if one of the twins has clinical pain,
0:12:15 there’s a very high chance that the identical twin will have it.
0:12:17 If they’re non identical, it’s lower,
0:12:21 and if they’re not related, then it falls way back.
0:12:25 So something like 50% is driven by genetic variances
0:12:27 that we get from our parents.
0:12:29 – You talk about phenotyping pain,
0:12:30 that’s what you mean essentially now
0:12:34 that you’re able to understand which type of pain
0:12:38 and therefore treat it, presumably treat it differently.
0:12:40 What was behind the scientific breakthroughs
0:12:42 that made that possible?
0:12:44 Were there technological reasons
0:12:47 or just sort of the accumulation over time of understanding?
0:12:50 What were the pinnacle moments that shifted things?
0:12:53 – Technology now does enable us to interrogate
0:12:55 the function of the nervous system
0:12:56 in a way that wasn’t possible.
0:13:01 We can see activity using a number of indicators.
0:13:03 We can also opt to genetically activate
0:13:05 specific parts of the nervous system.
0:13:09 It is now possible in preclinical models
0:13:13 to genetically target a light-sensitive protein
0:13:15 into defined sets of neurons,
0:13:17 and you can then use a laser light
0:13:19 to activate those sets of neurons
0:13:20 in a very controlled way.
0:13:22 – Super precise.
0:13:23 – Absolutely precise.
0:13:26 And you can then dissect out the rule
0:13:28 of each sets of neurons and particular circuits
0:13:30 in a way that just wasn’t possible.
0:13:33 So this is a very exciting technological breakthrough.
0:13:35 And the same thing with recording.
0:13:38 We can genetically put calcium-sensitive dyes
0:13:43 into specific sets of neurons and use little microscopes,
0:13:47 and from that can see the activity of broad populations
0:13:49 of neurons in different circumstances.
0:13:53 So this is an extremely exciting time in neuroscience,
0:13:56 but these technological breakthroughs
0:13:59 are enabling us to dissect out the function of the brain.
0:14:01 – So what you’re saying is essentially
0:14:02 with these new tools,
0:14:06 we’re seeing different areas light up in different ways
0:14:07 than we once expected,
0:14:11 which categorizes into different types of pain and reaction.
0:14:13 – Right, the breakthrough came when we realized
0:14:16 that acute nociceptive protective pain
0:14:19 was different from inflammatory or neuropathic,
0:14:22 and then what we call dysfunctional pain.
0:14:23 There’s been a whole set of data
0:14:27 looking at the mechanisms that drive each of them.
0:14:30 What exactly happens when you have a noxious pinch?
0:14:32 Which sets of neurons are activated?
0:14:35 How does this environmental mechanical stimulus
0:14:37 get converted into electrical activity?
0:14:39 Which neurons transmitted?
0:14:41 What are the signals that they use
0:14:45 to transfer the input from the periphery
0:14:46 to the central nervous system?
0:14:48 Which parts of the brain do they activate?
0:14:49 Which circuits?
0:14:51 All of that, we’re beginning to dissect out,
0:14:52 and as we do that,
0:14:54 we’re revealing potential points
0:14:57 at which there could be therapeutic intervention.
0:14:59 I think our greatest understanding now
0:15:01 is the protected pain.
0:15:04 How it switched on, which sets of neurons are responsible?
0:15:09 Which parts of the spinal cord and brain are involved?
0:15:12 We also know that in the presence of inflammation,
0:15:13 as part of the tissue injury,
0:15:16 there’s a massive recruitment of the immune system.
0:15:20 These produce signaling molecules that act on the neurons,
0:15:23 but there are also changes in the central nervous system as well.
0:15:25 As it becomes activated,
0:15:29 there are changes such that the transmission of input
0:15:32 can be amplified, it can spread,
0:15:34 so that, yes, at the site of the tissue injury,
0:15:37 there’s, as expected,
0:15:38 but actually the tissue surrounding it,
0:15:42 which is not injured, that is also hypersensitive,
0:15:44 and that is a manifestation of the presence
0:15:47 in the central nervous system
0:15:48 of what we call central sensitization,
0:15:50 an abnormal amplification.
0:15:51 So it’s happening both at the same time.
0:15:53 There’s both peripheral sensitization,
0:15:56 where the threshold is reduced,
0:15:58 and within the central nervous system,
0:16:01 there is an amplification of the signals
0:16:04 such that inputs that would normally not be painful
0:16:06 now begin to drive pain.
0:16:09 So now it begins to seem like our ways of treating pain
0:16:12 have been just this huge hammer
0:16:15 that we’ve been hitting all kinds of different things
0:16:17 with this giant blunt instrument.
0:16:18 – Absolutely.
0:16:21 – So let’s talk about the link between pain and addiction.
0:16:24 We think about pain as being linked to addiction
0:16:29 because of the category of opiates, largely.
0:16:32 Can I ask a little bit about the science behind
0:16:34 what we understand to be going on there,
0:16:36 neurologically speaking?
0:16:39 Why those two tend to go hand in hand,
0:16:41 or are we able to pull them apart?
0:16:43 – The big problem with opioids is that in addition
0:16:47 to reducing pain, they produce a euphoria.
0:16:50 The addiction comes from that euphoric signal,
0:16:52 which occurs in the brain in a different part of the brain
0:16:55 from where they act to reduce pain.
0:16:57 And so you have two parallel systems.
0:17:00 You have pain, and it’s a reduction by opioids.
0:17:03 And then you have the pleasure centers
0:17:05 that are activated by morphine,
0:17:07 which gives someone a high.
0:17:11 And not everyone has it, but those people who do,
0:17:13 then initiate a craving behavior.
0:17:16 And again, there are certain individuals
0:17:18 who have a high risk of craving,
0:17:20 who then become dependent and addicted.
0:17:25 But analgesia does not mean that there’s gonna be euphoria.
0:17:27 – It just so happens in that one.
0:17:29 – It just so happens in the case of opioids,
0:17:32 of which morphine is the typical example.
0:17:34 Not only does it reduce pain,
0:17:36 but it produces the sense of well-being,
0:17:40 of euphoria, of happiness.
0:17:42 But you don’t need to switch that system on
0:17:44 to get the pain relief.
0:17:47 So in a sense, it’s extremely bad luck.
0:17:49 The intertwining of pain and addiction
0:17:51 now is largely through the fact
0:17:54 that through opioid prescription
0:17:56 has until very recently been the means
0:17:59 by which people are introduced to opioids.
0:18:03 So until literally a few years ago,
0:18:05 if you had dental surgery,
0:18:09 your dentist would give you 20 oxycodone.
0:18:12 Why was it always so much?
0:18:17 It always seemed like much more than one would need.
0:18:21 Because what a big switch from your post-surgery beds
0:18:23 where they, okay, well, pain’s just part of it.
0:18:24 – Well, so basically,
0:18:27 this is not the first opioid crisis that we’ve had.
0:18:32 We’ve, in the 19th century, there was massive addiction,
0:18:33 the famous opium wars.
0:18:38 So the addictive qualities in the respiratory depression
0:18:41 and all the other bad features of opioids
0:18:45 have been around since the extract of the poppy was recognized.
0:18:47 It was for that reason
0:18:48 that in the middle of the 20th century,
0:18:50 physicians were wary,
0:18:53 highly wary of treating their patients with opioids.
0:18:56 And that’s why on my first exposure,
0:18:59 there were patients literally crying in pain
0:19:01 because of the surgeons at that time.
0:19:02 – Because of the backlash.
0:19:03 – Because of the backlash.
0:19:05 – Oh, how interesting, it’s like a pendulum.
0:19:09 So then what happened in the early 80s,
0:19:14 the view came that if you gave a morphine
0:19:17 and it’s equivalent drugs to patients in pain,
0:19:20 there was no risk of addiction.
0:19:22 This came from palliative care
0:19:26 because in the 80s, a growing recognition
0:19:28 that patients with terminal cancer
0:19:31 did need very active treatment.
0:19:35 And a big part of that clearly is morphine.
0:19:38 Those patients do need opioids,
0:19:41 most of them in the late stage of life.
0:19:43 And from that very good use,
0:19:44 it spread out to say,
0:19:47 well, patients who don’t have terminal cancer
0:19:50 who have pain also need opioids.
0:19:54 And a belief appeared in the medical profession
0:19:58 that if you gave morphine or any other opioid
0:20:01 to patients with pain, there was no risk of addiction.
0:20:03 There was not based on proper studies
0:20:05 and it’s turned out to be incorrect.
0:20:09 So that led to an enormous expansion
0:20:12 and the prescription of opioids.
0:20:14 – So it wasn’t so much that there was
0:20:16 any kind of intentional sort of extra,
0:20:17 it was just, it was thought not to matter,
0:20:19 not to pose a risk.
0:20:20 – And not to pose a risk.
0:20:22 It was done with the best of intentions.
0:20:25 So the idea was if you have a patient with pain,
0:20:27 give them an appropriate treatment.
0:20:31 So the prescription opioids were the avenue initially
0:20:33 in our current opioid crisis.
0:20:37 The means by which people could become introduced to opioids.
0:20:39 Although there’ve been many attempts
0:20:43 to try and design drugs that only have analgesia.
0:20:46 If they act through the mu-opioid receptor,
0:20:49 I think in the end the risks are very high
0:20:53 that you will always get the risk of addiction and abuse.
0:20:57 – But why is the euphoria addictive?
0:21:00 I mean, I feel euphoria, you know, over,
0:21:02 I don’t know, emotional events.
0:21:06 And I feel that I can experience it and let it go.
0:21:09 What is happening there that that kind of euphoria
0:21:12 leads to the uncontrolled craving?
0:21:15 – If you gave 100 people morphine,
0:21:18 some of them would feel terrible nausea.
0:21:20 Some would feel they are floating on a cloud
0:21:23 and happy and all their worries had disappeared
0:21:24 and others wouldn’t.
0:21:28 And that again reflects out the genetic variation.
0:21:30 But there are a group of people who,
0:21:35 when they are exposed to morphine, feel so pleasant.
0:21:38 And on top of it, when the morphine wears off,
0:21:42 they feel so terrible that they develop that craving.
0:21:45 So it’s a combination of the positive feeling they get
0:21:47 when they have the morphine
0:21:51 and the terrible negative feeling when it wears off.
0:21:54 It’s actually the strength of a negative feeling
0:21:58 that is one of the major drivers of addiction.
0:22:01 If you ask addicts, they’ll say their first exposure
0:22:03 was intensely pleasant
0:22:08 and they’ve never been able to reach that same high again.
0:22:11 However, when they stop it, they feel so terrible,
0:22:15 worse than they ever did starting off with,
0:22:16 that they now just trying to get back
0:22:19 to a normal balance feeling.
0:22:22 – Now I sort of understand the confluence of cultural
0:22:24 and what you grow up around
0:22:26 or how whole of a person you feel you are, right?
0:22:28 The psychological and the cultural stuff,
0:22:30 it’s about the contrast.
0:22:32 It’s not necessarily the euphoria,
0:22:36 but the contrast of the euphoria and the lack then of it.
0:22:39 – If you have a job that is not satisfying,
0:22:44 that if your family situation or your economic situation
0:22:46 is dire and all these other features,
0:22:49 then if you take a drug that makes you feel happy
0:22:54 and then that becomes a driver of that.
0:22:56 – So when you start to get this level of detail
0:23:00 and resolution sort of understanding the different pathways
0:23:03 and the different systems that are being triggered,
0:23:07 what do you see going forward for where we intervene?
0:23:09 Are there options starting to change
0:23:11 how we treat the pain as well?
0:23:14 – Absolutely, and I’m very optimistic.
0:23:17 I think we’re gonna move away from the blunt approach
0:23:19 where we see pain as a single box
0:23:22 and we throw only the same therapy at it.
0:23:26 To now having the means to identify in a patient,
0:23:27 what is the drive of their pain?
0:23:31 What is their particular set of pathological changes
0:23:33 that are causing their pain?
0:23:36 And that will suggest what is the most appropriate treatment?
0:23:40 And that’s called, the phenotype is the total pain
0:23:43 presentation in the patient and all its features
0:23:46 that will reflect the mechanisms that are driving it.
0:23:49 And then we can use that to what we call segmentation.
0:23:52 We can instead of seeing the population as a whole,
0:23:53 we can break it down.
0:23:56 This is a patient where the pain is entirely periphery.
0:23:59 It’s been driven by, for example,
0:24:02 if you had a thorn in your heel
0:24:04 and you didn’t remove it,
0:24:06 whenever you walked, you’d have pain.
0:24:09 But that pain has a definable cause
0:24:12 and in fact, all you need to do is remove that thorn
0:24:14 and your pain will disappear.
0:24:16 – Well, then you’ll have another kind of pain
0:24:17 for a little while, right?
0:24:19 – Absolutely, you can identify the cause
0:24:21 and you can target your treatment.
0:24:22 What the challenge for us now
0:24:25 is to find the equivalent of the thorn.
0:24:28 These are changes in the function of the nervous system.
0:24:31 But once we can find the mechanisms that are operating
0:24:33 and use that to help us choose
0:24:36 the most appropriate treatment for the patient,
0:24:38 one of the big problems in pain management
0:24:43 has been even those drugs that the FDA has approved,
0:24:48 such as pregabalin for treatment of neuropathic pain,
0:24:50 the vast majority of patients,
0:24:54 literally greater than 50% do not respond at all.
0:24:56 They get no benefit whatsoever.
0:24:59 – Was the drug chosen because of its lack
0:25:01 of side effects like addiction?
0:25:04 – So the FDA has two choices to make
0:25:05 when they are approving a drug.
0:25:07 The biggest factor is its safety.
0:25:12 So if a drug has no minimal adverse effects
0:25:15 or side effects, that is a very big plus.
0:25:17 Then they look for efficacy.
0:25:19 You need that in order to get approval
0:25:22 and put a label to say I can use this drug
0:25:24 to treat neuropathic pain.
0:25:26 But they don’t require you to say
0:25:29 it’ll work in all patients with neuropathic pain.
0:25:31 You just need to, the pharmaceutical company
0:25:34 just needs to show that it is significantly better
0:25:35 than no treatment.
0:25:40 And that may mean literally it works in 30%.
0:25:41 What about the 70%?
0:25:42 They get no benefit.
0:25:44 – But in the 30% that it did help,
0:25:46 was that a drug that intervened
0:25:49 in the pathways in a different way?
0:25:53 – It’s a case of serendipity or empirical observation.
0:25:57 Pre-gabalin was a derivative of an earlier drug
0:25:58 called gabapentin.
0:26:01 Gabapentin, as its name indicated,
0:26:04 was designed to be similar
0:26:09 to one of the major inhibitory transmitters in the brain.
0:26:11 It was designed in this way
0:26:15 to switch off overactivity of neurons in the brain
0:26:17 in the case of epilepsy and pain
0:26:18 and other similar conditions.
0:26:22 But it turned out that it didn’t actually act
0:26:23 in this way at all,
0:26:25 but it did have some analgesic signal.
0:26:29 And even after it was marketed and approved by the FDA,
0:26:32 it took a long time before it was discovered exactly
0:26:33 what that signal was.
0:26:34 And even to this day,
0:26:37 it’s not absolutely clear why activation
0:26:41 of those particular targets produces analgesia
0:26:43 in some patients and not in others.
0:26:45 – So how do you even begin to consider
0:26:47 new therapeutic approaches when you’re at the stage
0:26:49 of just sort of mapping?
0:26:51 Or do you feel we’re past mapping
0:26:52 and we’re starting to understand
0:26:54 other kinds of interventions?
0:26:56 What are the modalities of treatment for pain
0:26:59 that we’re gonna start beginning to see?
0:27:01 – A big part of my career has been trying
0:27:05 to identify the mechanisms of pain as a basis
0:27:08 for designing a new generation of treatments
0:27:11 that would mean that we could make opioids obsolete.
0:27:14 The technological advances in biomedicine as a whole
0:27:17 have really transformed the way we can approach
0:27:20 even something as complicated as pain.
0:27:24 We now know much more about the nature of tissue injury
0:27:25 and the inflammation that occurs
0:27:28 and the signaling molecules that are released
0:27:29 and how to block them.
0:27:33 We now know much more about the excitability of neurons
0:27:36 and how they change and how we can target that.
0:27:39 Whose signaling leads to the production of pain
0:27:41 may be different from others
0:27:43 and that enables us to get selectivity.
0:27:46 But we also, as I’ve indicated,
0:27:50 can recognize that in patients who have neuropathic pain,
0:27:53 we now know the downstream effects of that damage.
0:27:56 What are the pathophysiological changes that have occurred
0:28:00 and now can design our treatment to intervene?
0:28:02 At the moment, most of that treatment
0:28:06 just switches off those changes, so it’s symptom control.
0:28:09 But in the relative near future,
0:28:12 we will also get disease modifying treatment
0:28:16 where we can recognize that if a surgeon damages a nerve
0:28:18 during the surgery, there is a high risk,
0:28:21 particularly in those people who have a genetic predisposition
0:28:24 for the development of chronic neuropathic pain
0:28:28 and we will be able to say this person is at very high risk.
0:28:30 Therefore, we can give a treatment
0:28:32 that will prevent the evolution
0:28:35 of those pathological processes in the brain
0:28:39 that will cause that persistent neuropathic pain.
0:28:42 – So now you’re shifting to kind of prevention model
0:28:43 from the very beginning,
0:28:45 whereas they almost, by definition,
0:28:48 pain seemed like there was no possibility for preventative
0:28:50 unless you live in a bubble, right?
0:28:53 That we, by nature, encounter this.
0:28:55 I wanted to ask you something a little bit wacky about pain,
0:28:57 which is if these are certain receptors
0:28:59 and certain pathways of different types
0:29:01 and we’re beginning to understand what happens
0:29:02 in these different systems,
0:29:06 do you believe in our psychological ability to manage pain?
0:29:08 Is that a real thing?
0:29:12 Are there any ideas around self-hypnosis or meditation
0:29:14 or what effect do those have on the brain
0:29:17 or are we just telling ourselves stories?
0:29:20 – In the end, what we feel is the net result
0:29:23 of the totality of our brain function.
0:29:26 And so if you have pain and you’re depressed,
0:29:28 those will be additive.
0:29:30 If you have pain and you are depressed
0:29:32 and this interferes with your sleep,
0:29:34 that will increase your pain.
0:29:36 So you can get into vicious cycles
0:29:38 of pain gets worse and worse
0:29:40 and therefore your mood changes.
0:29:44 And therefore interventions such as distractions
0:29:47 may not switch off your pain, but can make it bearable.
0:29:50 And that certainly can be a big feature
0:29:53 and should be part of the totality of treatment.
0:29:56 We shouldn’t just rely on medication.
0:29:58 It’s generally not enough.
0:30:01 One of the breakthroughs over the last decade or so
0:30:03 has been the recognition that placebo,
0:30:05 we see that as a problem
0:30:06 because it makes it difficult
0:30:09 when we do clinical trials to identify new treatments.
0:30:14 In other words, you give a patient just a sugar pill
0:30:16 and they respond well to that
0:30:17 and how are you gonna differentiate that
0:30:19 from an active compound?
0:30:23 But that placebo is not wishful thinking.
0:30:26 It is the activation and we know that from functional imaging
0:30:29 of very specific parts of the brain
0:30:33 which then suppress or reduce the activity
0:30:34 of those parts of the brain
0:30:36 that generate the sensation of pain.
0:30:38 – So it is having a biological effect.
0:30:40 – It is a biological effect
0:30:42 and certainly anything that can switch it on
0:30:44 would be is beneficial.
0:30:46 It’s more variable and it’s more difficult to manage
0:30:49 but interventions such as acupuncture
0:30:51 are very good at switching on the placebo.
0:30:52 – That’s fascinating.
0:30:56 What you’re saying is that that actually could in turn
0:31:00 cycle back into how you experience the physical pain
0:31:02 just that in the overall composite
0:31:06 of how your brain as a system is behaving.
0:31:07 – There’s been a lot of work
0:31:10 in this trying to identify all these confounders
0:31:13 or contributors to pain.
0:31:16 And the one that seems strongest is something
0:31:18 what we call catastrophization.
0:31:22 That this is a definable feature of our makeup
0:31:26 that there are some people who if they have a problem
0:31:27 make it worse.
0:31:28 – It’s a kind of anxiety, right?
0:31:30 – It is exactly that.
0:31:33 And that definitely is highly correlated
0:31:37 with the intensity of pain that people experience.
0:31:39 And if you can target that catastrophization
0:31:42 where you can get people to overcome that anxiety,
0:31:44 the sense of powerlessness
0:31:46 and lack of ownership of that pain,
0:31:50 then that is again a very good treatment mentality.
0:31:51 – It’s not a magic switch.
0:31:54 It’s a cycle that you’re interrupting.
0:31:55 – And what needs to see in the context,
0:31:57 it’s more useful in the setting
0:32:02 if you have daily pain that you can control
0:32:04 and suppress and deal with.
0:32:06 – In the next five years,
0:32:08 how does our current framework need to shift
0:32:10 in the system again,
0:32:13 in order to allow this different kind of understanding
0:32:15 of pain to have actual effect in the clinic
0:32:16 and in the hospital?
0:32:18 – The greatest challenge for pain
0:32:21 is that it is a subjective experience.
0:32:24 I can’t know your pain, you can’t know mine.
0:32:28 When I say my pain is nine out of 10 on a 0 to 10 scale,
0:32:30 and tomorrow I say it’s five out of 10,
0:32:33 how accurate do you think that is?
0:32:35 – There’s a smiley faces in the picture, yeah.
0:32:37 – And it’s extremely difficult.
0:32:41 And what if the patient is a neonatal baby
0:32:45 or an adult who has Alzheimer’s disease
0:32:47 who cannot report their pain?
0:32:49 How do we measure that?
0:32:54 So I think a crucial element is we need new objective ways
0:32:57 of surrogates of pain, pain biomarkers.
0:33:00 One of the most promising ways, there’ll be others for sure,
0:33:02 but is to use functional brain imaging
0:33:05 to look at the changes in the activity patterns
0:33:06 in the brain.
0:33:07 This has been around for some time,
0:33:10 the signals have been very noisy,
0:33:13 but using artificial intelligence as a tool
0:33:15 to help us measure pain.
0:33:19 Now, instead of having a human look at the patterns
0:33:20 of changes of activity and say,
0:33:23 ah, that constitutes the presence of pain,
0:33:26 machine learning algorithms that are picking up patterns
0:33:28 that we wouldn’t even have detected.
0:33:29 – And correlating them to pain.
0:33:32 – And correlating, so it is reaching a point now
0:33:35 that the signals we can get from this functional image
0:33:38 will tell us that there’s a very high likelihood
0:33:41 that the patient has pain or not pain.
0:33:43 That also raises ethical issues,
0:33:44 because if someone says they have pain
0:33:48 and your functional imaging says there’s no equivalent,
0:33:49 how do you deal with that?
0:33:53 But the positive is that we are now have tools
0:33:56 to go beyond just relying on someone saying,
0:33:59 I have very bad pain or I have no pain,
0:34:00 to a point at which we can understand
0:34:03 the changes that underlie that pain
0:34:05 and the associated features.
0:34:08 Because not only do you pick up the sensory experience,
0:34:11 but you do pick up the changes that reflect the anxiety,
0:34:14 the helplessness, the other features
0:34:17 that constitute the entire picture
0:34:19 that a patient has.
0:34:20 – And to take it from the realm
0:34:23 of purely subjective to objective.
0:34:25 – Exactly, and that’s helping us
0:34:26 in our pre-clinical studies,
0:34:30 because again, pain is extremely complex,
0:34:31 there are many different features.
0:34:36 If we study those neurons that are the prime initiators
0:34:40 for pain and we can actually grow using stem cell technology,
0:34:42 we can take patient stem cells,
0:34:44 we can convert them into neurons,
0:34:47 we can look at features of the function of those neurons.
0:34:49 In the past, we used to look at one feature,
0:34:52 say the firing of action potentials
0:34:54 as a surrogate for their activity.
0:34:57 Now we recognize there are tens of features
0:35:00 of phenotypes as we call that constitute
0:35:02 the entire functional repertoire of these cells.
0:35:06 And again, machine learning is helping us see the patterns
0:35:09 and how they correlate with a normal state,
0:35:11 a disease state, what kind of disease state,
0:35:13 and the response to treatment.
0:35:17 So we now have tools at last to embrace complexity
0:35:21 and from that define what are the components
0:35:23 and which ones are relevant and important
0:35:25 and which ones to go for.
0:35:26 – And where to go next.
0:35:27 – And where to go next.
0:35:29 We are identifying other targets
0:35:30 that are looking very promising
0:35:35 and I’m optimistic that in consequence,
0:35:38 I think we are going to be able to make choices
0:35:40 about which are the best targets to go for,
0:35:43 find the means to identify the best modalities
0:35:46 to go for those targets,
0:35:49 to develop new ways of running clinical trials
0:35:50 that would be more sensitive,
0:35:52 both to suppress symptoms,
0:35:56 but also to prevent the evolution of permanent changes
0:35:57 in the function of the nervous system
0:36:00 that is a big driver of the chronicity of pain.
0:36:03 – Do you think that if we are able to get to that stage,
0:36:05 it would just roll through the system
0:36:08 and become adopted or would there be challenges?
0:36:10 – There are always challenges.
0:36:12 As is typical in biotech,
0:36:14 there are going to be many failures.
0:36:18 I think it is important that we go for risky
0:36:20 because some of them will pan out
0:36:24 but the portfolio of potential possibilities
0:36:26 is sufficiently exciting now
0:36:29 that I do think that we’ll be able to look back
0:36:31 in the near future and say,
0:36:34 at last we both understand pain
0:36:36 and we can now control it
0:36:39 in a way that has just simply not been possible before.
0:36:42 – I feel like we need a new thing in the doctor’s office
0:36:44 that says you’re experiencing one of these
0:36:48 four different types of pain for starters.
0:36:51 Thank you so much for joining us on the A16Z podcast.
0:36:51 – It’s a pleasure.
0:00:04 I’m Hannah.
0:00:07 In this episode, we talk all about the biology of pain
0:00:09 with Clifford Wolfe, professor of neurobiology
0:00:10 at Harvard Medical School,
0:00:13 who has dedicated his career to the subject.
0:00:15 The conversation covers everything you ever wondered
0:00:17 about the nature of pain, from why we even feel it,
0:00:20 and what the biological purpose of pain is,
0:00:22 to whether or not all pain is the same,
0:00:25 or if some kinds of pain function differently,
0:00:27 feel different, have different purposes,
0:00:30 or even if some people feel pain differently from others.
0:00:33 Wolfe describes the four different broad types
0:00:36 of pain we experience, what the purpose of each one is,
0:00:38 what it means now that we can phenotype
0:00:41 different kinds of pain and begin to understand them
0:00:43 as distinct and different from each other,
0:00:46 and how technology is enabling this new, deeper,
0:00:48 and much more complex understanding
0:00:50 of the nature of pain.
0:00:52 We also talk about what the biological link
0:00:54 between pain and addiction really is,
0:00:56 plus the one thing you might do
0:01:00 that we know makes your experience of pain worse.
0:01:01 You’ve been working on the subject of pain
0:01:03 for your entire career.
0:01:04 What drew you to that subject to begin with?
0:01:07 What was it that first sparked that initial interest for you
0:01:09 and where you began investigating?
0:01:11 – When I was a medical student,
0:01:15 I was on the surgical ward and at that time,
0:01:18 there was minimal treatment for postoperative pain,
0:01:21 so I came into the ward and there were a group of patients
0:01:24 who had major surgery,
0:01:28 and all of them were in severe discomfort,
0:01:31 and I said to the attending surgeon,
0:01:31 “What are you doing?
0:01:33 “Why aren’t you treating them with pain?”
0:01:35 He looked at me as if I was crazy,
0:01:36 and he said, “They’ve just had surgery.
0:01:38 “What do you expect? They’re in pain.”
0:01:41 – Even the idea of treating the pain was not–
0:01:42 – Absolutely, I was just thought,
0:01:43 “You have surgery, you have pain,”
0:01:45 and then you put up with it until the pain goes,
0:01:48 and I just thought to myself,
0:01:49 “That doesn’t sound right.”
0:01:51 At that time, there was very little understanding
0:01:53 of the mechanisms of pain,
0:01:56 and indeed, very few treatment options.
0:02:00 Now, as we have much greater understanding,
0:02:02 but unfortunately, the treatment options
0:02:03 haven’t really expanded,
0:02:08 and they are also accompanied by undesired effects,
0:02:10 particularly in the case of the opioids.
0:02:12 – So let’s talk a little bit about
0:02:14 what the understanding looked like then,
0:02:16 and what it’s beginning to look like now.
0:02:17 So what was our understanding
0:02:19 of the mechanism of pain back then
0:02:20 when we just sort of assumed,
0:02:22 “Okay, you have surgery, you’re in pain?”
0:02:25 – At that time, I just thought that all pain is similar,
0:02:27 that you either have no pain or you have pain,
0:02:31 and if you have pain, it’s an unpleasant sensation.
0:02:34 The closest analogy would be like flicking a switch
0:02:38 that switched on the pain sensation in your brain.
0:02:39 – So we thought it was that simple.
0:02:41 – We thought there was a peripheral trigger
0:02:43 which could be mild, like a pinprick,
0:02:45 or touching something too hard or too cold,
0:02:50 or larger in the sense of major trauma or post-surgical.
0:02:53 There was something that was activating that,
0:02:57 the pain switch, and that all the triggers of pain
0:02:59 acted on a single system in a single way,
0:03:02 and therefore pain really was not present,
0:03:04 or mild, moderate, or severe.
0:03:08 And that was the full range of the understanding of pain.
0:03:09 – Describe a little bit what the thought was
0:03:10 around that system.
0:03:12 Like how would that system be activated,
0:03:15 or what was our understanding of that system?
0:03:17 – Well, it certainly was appreciated
0:03:19 that pain has two facets.
0:03:24 One, it is an essential early warning device.
0:03:27 It tells us of danger in the environment.
0:03:30 And without it, there’s a high risk
0:03:31 that we may injure ourselves.
0:03:34 So you need this alarm system, if you like.
0:03:35 – And in the case of surgery,
0:03:38 that alarm system is going because you’re vulnerable,
0:03:39 you’re exposed, you’re healing.
0:03:42 – The alarm system is activated by the surgeon’s damage
0:03:45 that is created in the patient.
0:03:48 And then the alarm system continues
0:03:51 until that tissue injury is repaired.
0:03:53 But what is happening in a patient
0:03:55 who has persistent pain for years and years and years?
0:03:59 Why is that pain alarm activated continuously?
0:04:01 And there really was no explanation at the time.
0:04:04 – So that’s where you began your research?
0:04:05 – Exactly.
0:04:07 And at that time, if you had mild pain,
0:04:09 then you took a non-steroidal anti-inflammatory drug
0:04:12 like Tylenol or Advil.
0:04:14 If you had more severe pain,
0:04:17 you added a weak opioid to your NSAID.
0:04:19 And if you had severe pain,
0:04:22 then that’s the point at which you’d give a stronger opioid.
0:04:26 And the World Health Organization had an analgesic ladder
0:04:29 that it literally was that mild, moderate, severe pain
0:04:31 increasing the strength of the analgesic.
0:04:33 And again, in keeping with this notion
0:04:35 that pain was a unitary system activated
0:04:38 by different triggers of different intensity.
0:04:39 – It’s so simple.
0:04:42 It makes me think of the signs in the nurse’s doctor’s office
0:04:43 with the smiling faces.
0:04:45 Like even that is a whole order of magnitude
0:04:47 above in terms of understanding.
0:04:51 – Unfortunately, the reality is those smiling faces
0:04:55 would not be smiling if they saw how complicated pain was.
0:04:59 The notion that there’s a simple switch is just incorrect.
0:05:02 The notion that pain can be defined by its severity
0:05:05 as mild, moderate, or severe is incorrect.
0:05:07 And the notion that the treatment should be based on that
0:05:09 is also incorrect.
0:05:11 – What is it that has shifted in our understanding
0:05:14 of the biology of how much more complex it is?
0:05:17 – Pain as an early warning device,
0:05:20 kind of radar for danger, if you like,
0:05:23 is completely different from the pain that occurs
0:05:26 in the setting of a patient who has persistent pain.
0:05:28 – The quality of the pain?
0:05:30 – No, that was the confusing element.
0:05:32 We use the same word pain to describe both
0:05:35 and therefore physicians and patients felt
0:05:36 this is the same phenomenon.
0:05:39 But in fact, the pain that we experience
0:05:43 does not reflect what is initiating and sustaining the pain.
0:05:47 In the case of pain as a protective mechanism,
0:05:49 you need just an intense stimulus,
0:05:52 one of sufficient intensity
0:05:55 that potentially can damage our tissues.
0:05:57 And we have this very elaborate system
0:06:00 that has been designed to detect that danger.
0:06:03 So we have specialized sensory neurons
0:06:06 that are activated only by intense stimuli.
0:06:08 And when they reach that threshold,
0:06:11 which is just below the point at which damage can occur,
0:06:15 we feel a pin prick or the point at which something
0:06:19 that is too warm and it has the risk of burning us.
0:06:21 – And is that threshold the same for everyone
0:06:22 or different for everyone?
0:06:26 – It’s surprisingly similar that if you measure it
0:06:31 and in a situation where you remove all the emotional
0:06:35 aspects of pain, the threshold at which someone feels
0:06:38 something has been warm and then hotter.
0:06:40 And then when it switches to actually painful,
0:06:45 is it 42 degrees with a surprisingly small variation?
0:06:46 – That is surprising.
0:06:50 So the idea of being more pain tolerant
0:06:54 because of your accumulated experiences or sensitivities,
0:06:55 is that not…
0:06:57 – That is true, but that is not for that kind of pain.
0:07:01 If you’re drinking your coffee and it’s too hot
0:07:02 and it burns your tongue,
0:07:04 you will stop drinking it immediately.
0:07:05 You weren’t even aware of it.
0:07:08 We’ve all learned to take a little sip
0:07:11 and maybe not even put the coffee in our mouth.
0:07:13 So we’ve been trained by our experience.
0:07:15 We try and avoid that pain.
0:07:17 For that protective pain to work,
0:07:19 the pain has to be unpleasant.
0:07:21 And in order for it to work,
0:07:22 you can’t say, I’ll deal with it later.
0:07:25 You’ve got to deal with it immediately.
0:07:28 So it demands our attention and it’s intense
0:07:30 and it’s unpleasant.
0:07:32 And in that sense, it is that analogy of a switch.
0:07:35 You have switched on something that focus your attention.
0:07:36 I have to do something.
0:07:39 I have to avoid this situation immediately.
0:07:41 We’re talking a matter of seconds.
0:07:42 – It’s actually incredibly effective
0:07:45 when you think about it like that, a very effective tool.
0:07:46 – The protective mechanism of pain
0:07:49 requires an intense stimulus to activate it,
0:07:50 a noxious stimulus.
0:07:53 And we actually call this nociceptive pain,
0:07:56 which is the pain that is initiated by a noxious stimuli.
0:07:58 Now, the big difference
0:08:00 coming back to our post-surgical pain,
0:08:03 we now no longer dealing with protection from injury
0:08:04 because the injuries occurred.
0:08:06 It could either occur dramatically
0:08:08 if you’re involved in a motor vehicle accident
0:08:10 or it could be because you have surgery.
0:08:12 In both cases, there’s trauma to your tissue.
0:08:15 So you no longer need a protective system
0:08:18 to prevent tissue damage because you have tissue damage.
0:08:20 So what happens there?
0:08:23 And instead of protecting you from tissue damage,
0:08:26 evolution has developed a system
0:08:29 that helps the injury to heal.
0:08:31 And how does it do that?
0:08:34 It makes the injured part hypersensitive.
0:08:37 If you think about an injury or an inflammation,
0:08:40 just touching that wounded site will produce pain.
0:08:43 So we’ve now got a complete shift.
0:08:46 The pain is now activated by a stimulus
0:08:48 that would normally be innocuous,
0:08:51 a sensation that would normally just be light touch,
0:08:53 or in the case of a bad sunburn,
0:08:57 a shower that would normally be pleasantly warm
0:08:58 is now burning hot,
0:09:01 even though the temperature is not in the hot range.
0:09:03 – So it’s not about the stimulus,
0:09:04 it’s about the condition.
0:09:05 – Absolutely.
0:09:06 The nervous system has changed.
0:09:08 It’s become hypersensitive,
0:09:11 such that normally innocuous stimuli
0:09:14 are now activating the pain system,
0:09:17 and we have a tenderness, soreness,
0:09:19 a heightened sense of pain.
0:09:21 This is an adaptive thing.
0:09:25 For example, after major abdominal surgery,
0:09:27 if there were a way to completely eliminate the pain
0:09:30 and you say, “Fine, I’ll go for a jog,”
0:09:31 that would not be a good idea
0:09:33 because you’ve got this wounded tissue,
0:09:37 it needs time for the wound to heal, to repair itself.
0:09:40 And the same with if you’ve got arthritis,
0:09:42 where there’s damage to the joint,
0:09:43 if you excessively use the joint,
0:09:46 if you were pain-free, you would damage that joint.
0:09:49 – It’s still essentially protective in nature.
0:09:52 – It’s not protecting you from the external environment
0:09:54 and potential danger to your tissue,
0:09:59 it’s protecting the injured body part from further injury,
0:10:01 allowing repair to occur.
0:10:03 And so, finally, the pain that I thought
0:10:05 I first wanted to cure, post-operative pain
0:10:07 or post-traumatic pain,
0:10:09 so you do want to try and reduce it
0:10:10 to make people feel comfortable,
0:10:12 but at the same time, if you eliminated,
0:10:14 that would be a real problem.
0:10:17 But that then leaves our patients
0:10:19 who may not have any surgery,
0:10:21 they have no tissue damage,
0:10:24 and who have pain that just persists
0:10:26 for years and years and years.
0:10:28 One of the commonest causes of that
0:10:30 is damage to the nervous system itself,
0:10:32 and we call that neuropathic pain.
0:10:36 And we now appreciate this is no longer telling us
0:10:38 about the presence of noxious stimuli
0:10:40 or the presence of an injured part,
0:10:43 it is the malfunction of the nervous system itself,
0:10:45 it is a disease of the nervous system,
0:10:48 and it’s mechanistically quite different
0:10:50 from either the two other kinds of pain.
0:10:53 – So there’s three major categories of pain now.
0:10:54 – And there’s one fourth one.
0:10:55 – Oh, really?
0:10:58 – And this is a group of patients
0:11:02 who tend not exclusively, but tend to be women,
0:11:04 there’s no noxious stimuli,
0:11:06 there’s no tissue inflammation,
0:11:08 there’s no damage to the nervous system,
0:11:12 but they may have chronic widespread pain.
0:11:15 And for a long time, this was labeled as psychosomatic,
0:11:18 and there was almost a blame feature
0:11:21 to that label to the patients.
0:11:25 But now we realize they too have an abnormal nervous system.
0:11:27 The reason they’re feeling pain
0:11:30 is because it is if someone has switched up
0:11:34 the volume control in their central nervous system,
0:11:37 such that what would normally be a soft sound
0:11:40 is now very loud and for them,
0:11:43 a soft touch now produces pain.
0:11:46 So this is a pathological functioning of the nervous system
0:11:49 in the absence of damage to the nervous system.
0:11:51 – Is it congenital in those cases?
0:11:54 – Overall, for most clinical pain conditions,
0:11:59 there is actually a very strong heritable component,
0:12:03 something like 50% of the risk of developing pain
0:12:04 is heritable.
0:12:05 We know that from twin studies.
0:12:08 So if you look at identical twins,
0:12:12 if one of the twins has clinical pain,
0:12:15 there’s a very high chance that the identical twin will have it.
0:12:17 If they’re non identical, it’s lower,
0:12:21 and if they’re not related, then it falls way back.
0:12:25 So something like 50% is driven by genetic variances
0:12:27 that we get from our parents.
0:12:29 – You talk about phenotyping pain,
0:12:30 that’s what you mean essentially now
0:12:34 that you’re able to understand which type of pain
0:12:38 and therefore treat it, presumably treat it differently.
0:12:40 What was behind the scientific breakthroughs
0:12:42 that made that possible?
0:12:44 Were there technological reasons
0:12:47 or just sort of the accumulation over time of understanding?
0:12:50 What were the pinnacle moments that shifted things?
0:12:53 – Technology now does enable us to interrogate
0:12:55 the function of the nervous system
0:12:56 in a way that wasn’t possible.
0:13:01 We can see activity using a number of indicators.
0:13:03 We can also opt to genetically activate
0:13:05 specific parts of the nervous system.
0:13:09 It is now possible in preclinical models
0:13:13 to genetically target a light-sensitive protein
0:13:15 into defined sets of neurons,
0:13:17 and you can then use a laser light
0:13:19 to activate those sets of neurons
0:13:20 in a very controlled way.
0:13:22 – Super precise.
0:13:23 – Absolutely precise.
0:13:26 And you can then dissect out the rule
0:13:28 of each sets of neurons and particular circuits
0:13:30 in a way that just wasn’t possible.
0:13:33 So this is a very exciting technological breakthrough.
0:13:35 And the same thing with recording.
0:13:38 We can genetically put calcium-sensitive dyes
0:13:43 into specific sets of neurons and use little microscopes,
0:13:47 and from that can see the activity of broad populations
0:13:49 of neurons in different circumstances.
0:13:53 So this is an extremely exciting time in neuroscience,
0:13:56 but these technological breakthroughs
0:13:59 are enabling us to dissect out the function of the brain.
0:14:01 – So what you’re saying is essentially
0:14:02 with these new tools,
0:14:06 we’re seeing different areas light up in different ways
0:14:07 than we once expected,
0:14:11 which categorizes into different types of pain and reaction.
0:14:13 – Right, the breakthrough came when we realized
0:14:16 that acute nociceptive protective pain
0:14:19 was different from inflammatory or neuropathic,
0:14:22 and then what we call dysfunctional pain.
0:14:23 There’s been a whole set of data
0:14:27 looking at the mechanisms that drive each of them.
0:14:30 What exactly happens when you have a noxious pinch?
0:14:32 Which sets of neurons are activated?
0:14:35 How does this environmental mechanical stimulus
0:14:37 get converted into electrical activity?
0:14:39 Which neurons transmitted?
0:14:41 What are the signals that they use
0:14:45 to transfer the input from the periphery
0:14:46 to the central nervous system?
0:14:48 Which parts of the brain do they activate?
0:14:49 Which circuits?
0:14:51 All of that, we’re beginning to dissect out,
0:14:52 and as we do that,
0:14:54 we’re revealing potential points
0:14:57 at which there could be therapeutic intervention.
0:14:59 I think our greatest understanding now
0:15:01 is the protected pain.
0:15:04 How it switched on, which sets of neurons are responsible?
0:15:09 Which parts of the spinal cord and brain are involved?
0:15:12 We also know that in the presence of inflammation,
0:15:13 as part of the tissue injury,
0:15:16 there’s a massive recruitment of the immune system.
0:15:20 These produce signaling molecules that act on the neurons,
0:15:23 but there are also changes in the central nervous system as well.
0:15:25 As it becomes activated,
0:15:29 there are changes such that the transmission of input
0:15:32 can be amplified, it can spread,
0:15:34 so that, yes, at the site of the tissue injury,
0:15:37 there’s, as expected,
0:15:38 but actually the tissue surrounding it,
0:15:42 which is not injured, that is also hypersensitive,
0:15:44 and that is a manifestation of the presence
0:15:47 in the central nervous system
0:15:48 of what we call central sensitization,
0:15:50 an abnormal amplification.
0:15:51 So it’s happening both at the same time.
0:15:53 There’s both peripheral sensitization,
0:15:56 where the threshold is reduced,
0:15:58 and within the central nervous system,
0:16:01 there is an amplification of the signals
0:16:04 such that inputs that would normally not be painful
0:16:06 now begin to drive pain.
0:16:09 So now it begins to seem like our ways of treating pain
0:16:12 have been just this huge hammer
0:16:15 that we’ve been hitting all kinds of different things
0:16:17 with this giant blunt instrument.
0:16:18 – Absolutely.
0:16:21 – So let’s talk about the link between pain and addiction.
0:16:24 We think about pain as being linked to addiction
0:16:29 because of the category of opiates, largely.
0:16:32 Can I ask a little bit about the science behind
0:16:34 what we understand to be going on there,
0:16:36 neurologically speaking?
0:16:39 Why those two tend to go hand in hand,
0:16:41 or are we able to pull them apart?
0:16:43 – The big problem with opioids is that in addition
0:16:47 to reducing pain, they produce a euphoria.
0:16:50 The addiction comes from that euphoric signal,
0:16:52 which occurs in the brain in a different part of the brain
0:16:55 from where they act to reduce pain.
0:16:57 And so you have two parallel systems.
0:17:00 You have pain, and it’s a reduction by opioids.
0:17:03 And then you have the pleasure centers
0:17:05 that are activated by morphine,
0:17:07 which gives someone a high.
0:17:11 And not everyone has it, but those people who do,
0:17:13 then initiate a craving behavior.
0:17:16 And again, there are certain individuals
0:17:18 who have a high risk of craving,
0:17:20 who then become dependent and addicted.
0:17:25 But analgesia does not mean that there’s gonna be euphoria.
0:17:27 – It just so happens in that one.
0:17:29 – It just so happens in the case of opioids,
0:17:32 of which morphine is the typical example.
0:17:34 Not only does it reduce pain,
0:17:36 but it produces the sense of well-being,
0:17:40 of euphoria, of happiness.
0:17:42 But you don’t need to switch that system on
0:17:44 to get the pain relief.
0:17:47 So in a sense, it’s extremely bad luck.
0:17:49 The intertwining of pain and addiction
0:17:51 now is largely through the fact
0:17:54 that through opioid prescription
0:17:56 has until very recently been the means
0:17:59 by which people are introduced to opioids.
0:18:03 So until literally a few years ago,
0:18:05 if you had dental surgery,
0:18:09 your dentist would give you 20 oxycodone.
0:18:12 Why was it always so much?
0:18:17 It always seemed like much more than one would need.
0:18:21 Because what a big switch from your post-surgery beds
0:18:23 where they, okay, well, pain’s just part of it.
0:18:24 – Well, so basically,
0:18:27 this is not the first opioid crisis that we’ve had.
0:18:32 We’ve, in the 19th century, there was massive addiction,
0:18:33 the famous opium wars.
0:18:38 So the addictive qualities in the respiratory depression
0:18:41 and all the other bad features of opioids
0:18:45 have been around since the extract of the poppy was recognized.
0:18:47 It was for that reason
0:18:48 that in the middle of the 20th century,
0:18:50 physicians were wary,
0:18:53 highly wary of treating their patients with opioids.
0:18:56 And that’s why on my first exposure,
0:18:59 there were patients literally crying in pain
0:19:01 because of the surgeons at that time.
0:19:02 – Because of the backlash.
0:19:03 – Because of the backlash.
0:19:05 – Oh, how interesting, it’s like a pendulum.
0:19:09 So then what happened in the early 80s,
0:19:14 the view came that if you gave a morphine
0:19:17 and it’s equivalent drugs to patients in pain,
0:19:20 there was no risk of addiction.
0:19:22 This came from palliative care
0:19:26 because in the 80s, a growing recognition
0:19:28 that patients with terminal cancer
0:19:31 did need very active treatment.
0:19:35 And a big part of that clearly is morphine.
0:19:38 Those patients do need opioids,
0:19:41 most of them in the late stage of life.
0:19:43 And from that very good use,
0:19:44 it spread out to say,
0:19:47 well, patients who don’t have terminal cancer
0:19:50 who have pain also need opioids.
0:19:54 And a belief appeared in the medical profession
0:19:58 that if you gave morphine or any other opioid
0:20:01 to patients with pain, there was no risk of addiction.
0:20:03 There was not based on proper studies
0:20:05 and it’s turned out to be incorrect.
0:20:09 So that led to an enormous expansion
0:20:12 and the prescription of opioids.
0:20:14 – So it wasn’t so much that there was
0:20:16 any kind of intentional sort of extra,
0:20:17 it was just, it was thought not to matter,
0:20:19 not to pose a risk.
0:20:20 – And not to pose a risk.
0:20:22 It was done with the best of intentions.
0:20:25 So the idea was if you have a patient with pain,
0:20:27 give them an appropriate treatment.
0:20:31 So the prescription opioids were the avenue initially
0:20:33 in our current opioid crisis.
0:20:37 The means by which people could become introduced to opioids.
0:20:39 Although there’ve been many attempts
0:20:43 to try and design drugs that only have analgesia.
0:20:46 If they act through the mu-opioid receptor,
0:20:49 I think in the end the risks are very high
0:20:53 that you will always get the risk of addiction and abuse.
0:20:57 – But why is the euphoria addictive?
0:21:00 I mean, I feel euphoria, you know, over,
0:21:02 I don’t know, emotional events.
0:21:06 And I feel that I can experience it and let it go.
0:21:09 What is happening there that that kind of euphoria
0:21:12 leads to the uncontrolled craving?
0:21:15 – If you gave 100 people morphine,
0:21:18 some of them would feel terrible nausea.
0:21:20 Some would feel they are floating on a cloud
0:21:23 and happy and all their worries had disappeared
0:21:24 and others wouldn’t.
0:21:28 And that again reflects out the genetic variation.
0:21:30 But there are a group of people who,
0:21:35 when they are exposed to morphine, feel so pleasant.
0:21:38 And on top of it, when the morphine wears off,
0:21:42 they feel so terrible that they develop that craving.
0:21:45 So it’s a combination of the positive feeling they get
0:21:47 when they have the morphine
0:21:51 and the terrible negative feeling when it wears off.
0:21:54 It’s actually the strength of a negative feeling
0:21:58 that is one of the major drivers of addiction.
0:22:01 If you ask addicts, they’ll say their first exposure
0:22:03 was intensely pleasant
0:22:08 and they’ve never been able to reach that same high again.
0:22:11 However, when they stop it, they feel so terrible,
0:22:15 worse than they ever did starting off with,
0:22:16 that they now just trying to get back
0:22:19 to a normal balance feeling.
0:22:22 – Now I sort of understand the confluence of cultural
0:22:24 and what you grow up around
0:22:26 or how whole of a person you feel you are, right?
0:22:28 The psychological and the cultural stuff,
0:22:30 it’s about the contrast.
0:22:32 It’s not necessarily the euphoria,
0:22:36 but the contrast of the euphoria and the lack then of it.
0:22:39 – If you have a job that is not satisfying,
0:22:44 that if your family situation or your economic situation
0:22:46 is dire and all these other features,
0:22:49 then if you take a drug that makes you feel happy
0:22:54 and then that becomes a driver of that.
0:22:56 – So when you start to get this level of detail
0:23:00 and resolution sort of understanding the different pathways
0:23:03 and the different systems that are being triggered,
0:23:07 what do you see going forward for where we intervene?
0:23:09 Are there options starting to change
0:23:11 how we treat the pain as well?
0:23:14 – Absolutely, and I’m very optimistic.
0:23:17 I think we’re gonna move away from the blunt approach
0:23:19 where we see pain as a single box
0:23:22 and we throw only the same therapy at it.
0:23:26 To now having the means to identify in a patient,
0:23:27 what is the drive of their pain?
0:23:31 What is their particular set of pathological changes
0:23:33 that are causing their pain?
0:23:36 And that will suggest what is the most appropriate treatment?
0:23:40 And that’s called, the phenotype is the total pain
0:23:43 presentation in the patient and all its features
0:23:46 that will reflect the mechanisms that are driving it.
0:23:49 And then we can use that to what we call segmentation.
0:23:52 We can instead of seeing the population as a whole,
0:23:53 we can break it down.
0:23:56 This is a patient where the pain is entirely periphery.
0:23:59 It’s been driven by, for example,
0:24:02 if you had a thorn in your heel
0:24:04 and you didn’t remove it,
0:24:06 whenever you walked, you’d have pain.
0:24:09 But that pain has a definable cause
0:24:12 and in fact, all you need to do is remove that thorn
0:24:14 and your pain will disappear.
0:24:16 – Well, then you’ll have another kind of pain
0:24:17 for a little while, right?
0:24:19 – Absolutely, you can identify the cause
0:24:21 and you can target your treatment.
0:24:22 What the challenge for us now
0:24:25 is to find the equivalent of the thorn.
0:24:28 These are changes in the function of the nervous system.
0:24:31 But once we can find the mechanisms that are operating
0:24:33 and use that to help us choose
0:24:36 the most appropriate treatment for the patient,
0:24:38 one of the big problems in pain management
0:24:43 has been even those drugs that the FDA has approved,
0:24:48 such as pregabalin for treatment of neuropathic pain,
0:24:50 the vast majority of patients,
0:24:54 literally greater than 50% do not respond at all.
0:24:56 They get no benefit whatsoever.
0:24:59 – Was the drug chosen because of its lack
0:25:01 of side effects like addiction?
0:25:04 – So the FDA has two choices to make
0:25:05 when they are approving a drug.
0:25:07 The biggest factor is its safety.
0:25:12 So if a drug has no minimal adverse effects
0:25:15 or side effects, that is a very big plus.
0:25:17 Then they look for efficacy.
0:25:19 You need that in order to get approval
0:25:22 and put a label to say I can use this drug
0:25:24 to treat neuropathic pain.
0:25:26 But they don’t require you to say
0:25:29 it’ll work in all patients with neuropathic pain.
0:25:31 You just need to, the pharmaceutical company
0:25:34 just needs to show that it is significantly better
0:25:35 than no treatment.
0:25:40 And that may mean literally it works in 30%.
0:25:41 What about the 70%?
0:25:42 They get no benefit.
0:25:44 – But in the 30% that it did help,
0:25:46 was that a drug that intervened
0:25:49 in the pathways in a different way?
0:25:53 – It’s a case of serendipity or empirical observation.
0:25:57 Pre-gabalin was a derivative of an earlier drug
0:25:58 called gabapentin.
0:26:01 Gabapentin, as its name indicated,
0:26:04 was designed to be similar
0:26:09 to one of the major inhibitory transmitters in the brain.
0:26:11 It was designed in this way
0:26:15 to switch off overactivity of neurons in the brain
0:26:17 in the case of epilepsy and pain
0:26:18 and other similar conditions.
0:26:22 But it turned out that it didn’t actually act
0:26:23 in this way at all,
0:26:25 but it did have some analgesic signal.
0:26:29 And even after it was marketed and approved by the FDA,
0:26:32 it took a long time before it was discovered exactly
0:26:33 what that signal was.
0:26:34 And even to this day,
0:26:37 it’s not absolutely clear why activation
0:26:41 of those particular targets produces analgesia
0:26:43 in some patients and not in others.
0:26:45 – So how do you even begin to consider
0:26:47 new therapeutic approaches when you’re at the stage
0:26:49 of just sort of mapping?
0:26:51 Or do you feel we’re past mapping
0:26:52 and we’re starting to understand
0:26:54 other kinds of interventions?
0:26:56 What are the modalities of treatment for pain
0:26:59 that we’re gonna start beginning to see?
0:27:01 – A big part of my career has been trying
0:27:05 to identify the mechanisms of pain as a basis
0:27:08 for designing a new generation of treatments
0:27:11 that would mean that we could make opioids obsolete.
0:27:14 The technological advances in biomedicine as a whole
0:27:17 have really transformed the way we can approach
0:27:20 even something as complicated as pain.
0:27:24 We now know much more about the nature of tissue injury
0:27:25 and the inflammation that occurs
0:27:28 and the signaling molecules that are released
0:27:29 and how to block them.
0:27:33 We now know much more about the excitability of neurons
0:27:36 and how they change and how we can target that.
0:27:39 Whose signaling leads to the production of pain
0:27:41 may be different from others
0:27:43 and that enables us to get selectivity.
0:27:46 But we also, as I’ve indicated,
0:27:50 can recognize that in patients who have neuropathic pain,
0:27:53 we now know the downstream effects of that damage.
0:27:56 What are the pathophysiological changes that have occurred
0:28:00 and now can design our treatment to intervene?
0:28:02 At the moment, most of that treatment
0:28:06 just switches off those changes, so it’s symptom control.
0:28:09 But in the relative near future,
0:28:12 we will also get disease modifying treatment
0:28:16 where we can recognize that if a surgeon damages a nerve
0:28:18 during the surgery, there is a high risk,
0:28:21 particularly in those people who have a genetic predisposition
0:28:24 for the development of chronic neuropathic pain
0:28:28 and we will be able to say this person is at very high risk.
0:28:30 Therefore, we can give a treatment
0:28:32 that will prevent the evolution
0:28:35 of those pathological processes in the brain
0:28:39 that will cause that persistent neuropathic pain.
0:28:42 – So now you’re shifting to kind of prevention model
0:28:43 from the very beginning,
0:28:45 whereas they almost, by definition,
0:28:48 pain seemed like there was no possibility for preventative
0:28:50 unless you live in a bubble, right?
0:28:53 That we, by nature, encounter this.
0:28:55 I wanted to ask you something a little bit wacky about pain,
0:28:57 which is if these are certain receptors
0:28:59 and certain pathways of different types
0:29:01 and we’re beginning to understand what happens
0:29:02 in these different systems,
0:29:06 do you believe in our psychological ability to manage pain?
0:29:08 Is that a real thing?
0:29:12 Are there any ideas around self-hypnosis or meditation
0:29:14 or what effect do those have on the brain
0:29:17 or are we just telling ourselves stories?
0:29:20 – In the end, what we feel is the net result
0:29:23 of the totality of our brain function.
0:29:26 And so if you have pain and you’re depressed,
0:29:28 those will be additive.
0:29:30 If you have pain and you are depressed
0:29:32 and this interferes with your sleep,
0:29:34 that will increase your pain.
0:29:36 So you can get into vicious cycles
0:29:38 of pain gets worse and worse
0:29:40 and therefore your mood changes.
0:29:44 And therefore interventions such as distractions
0:29:47 may not switch off your pain, but can make it bearable.
0:29:50 And that certainly can be a big feature
0:29:53 and should be part of the totality of treatment.
0:29:56 We shouldn’t just rely on medication.
0:29:58 It’s generally not enough.
0:30:01 One of the breakthroughs over the last decade or so
0:30:03 has been the recognition that placebo,
0:30:05 we see that as a problem
0:30:06 because it makes it difficult
0:30:09 when we do clinical trials to identify new treatments.
0:30:14 In other words, you give a patient just a sugar pill
0:30:16 and they respond well to that
0:30:17 and how are you gonna differentiate that
0:30:19 from an active compound?
0:30:23 But that placebo is not wishful thinking.
0:30:26 It is the activation and we know that from functional imaging
0:30:29 of very specific parts of the brain
0:30:33 which then suppress or reduce the activity
0:30:34 of those parts of the brain
0:30:36 that generate the sensation of pain.
0:30:38 – So it is having a biological effect.
0:30:40 – It is a biological effect
0:30:42 and certainly anything that can switch it on
0:30:44 would be is beneficial.
0:30:46 It’s more variable and it’s more difficult to manage
0:30:49 but interventions such as acupuncture
0:30:51 are very good at switching on the placebo.
0:30:52 – That’s fascinating.
0:30:56 What you’re saying is that that actually could in turn
0:31:00 cycle back into how you experience the physical pain
0:31:02 just that in the overall composite
0:31:06 of how your brain as a system is behaving.
0:31:07 – There’s been a lot of work
0:31:10 in this trying to identify all these confounders
0:31:13 or contributors to pain.
0:31:16 And the one that seems strongest is something
0:31:18 what we call catastrophization.
0:31:22 That this is a definable feature of our makeup
0:31:26 that there are some people who if they have a problem
0:31:27 make it worse.
0:31:28 – It’s a kind of anxiety, right?
0:31:30 – It is exactly that.
0:31:33 And that definitely is highly correlated
0:31:37 with the intensity of pain that people experience.
0:31:39 And if you can target that catastrophization
0:31:42 where you can get people to overcome that anxiety,
0:31:44 the sense of powerlessness
0:31:46 and lack of ownership of that pain,
0:31:50 then that is again a very good treatment mentality.
0:31:51 – It’s not a magic switch.
0:31:54 It’s a cycle that you’re interrupting.
0:31:55 – And what needs to see in the context,
0:31:57 it’s more useful in the setting
0:32:02 if you have daily pain that you can control
0:32:04 and suppress and deal with.
0:32:06 – In the next five years,
0:32:08 how does our current framework need to shift
0:32:10 in the system again,
0:32:13 in order to allow this different kind of understanding
0:32:15 of pain to have actual effect in the clinic
0:32:16 and in the hospital?
0:32:18 – The greatest challenge for pain
0:32:21 is that it is a subjective experience.
0:32:24 I can’t know your pain, you can’t know mine.
0:32:28 When I say my pain is nine out of 10 on a 0 to 10 scale,
0:32:30 and tomorrow I say it’s five out of 10,
0:32:33 how accurate do you think that is?
0:32:35 – There’s a smiley faces in the picture, yeah.
0:32:37 – And it’s extremely difficult.
0:32:41 And what if the patient is a neonatal baby
0:32:45 or an adult who has Alzheimer’s disease
0:32:47 who cannot report their pain?
0:32:49 How do we measure that?
0:32:54 So I think a crucial element is we need new objective ways
0:32:57 of surrogates of pain, pain biomarkers.
0:33:00 One of the most promising ways, there’ll be others for sure,
0:33:02 but is to use functional brain imaging
0:33:05 to look at the changes in the activity patterns
0:33:06 in the brain.
0:33:07 This has been around for some time,
0:33:10 the signals have been very noisy,
0:33:13 but using artificial intelligence as a tool
0:33:15 to help us measure pain.
0:33:19 Now, instead of having a human look at the patterns
0:33:20 of changes of activity and say,
0:33:23 ah, that constitutes the presence of pain,
0:33:26 machine learning algorithms that are picking up patterns
0:33:28 that we wouldn’t even have detected.
0:33:29 – And correlating them to pain.
0:33:32 – And correlating, so it is reaching a point now
0:33:35 that the signals we can get from this functional image
0:33:38 will tell us that there’s a very high likelihood
0:33:41 that the patient has pain or not pain.
0:33:43 That also raises ethical issues,
0:33:44 because if someone says they have pain
0:33:48 and your functional imaging says there’s no equivalent,
0:33:49 how do you deal with that?
0:33:53 But the positive is that we are now have tools
0:33:56 to go beyond just relying on someone saying,
0:33:59 I have very bad pain or I have no pain,
0:34:00 to a point at which we can understand
0:34:03 the changes that underlie that pain
0:34:05 and the associated features.
0:34:08 Because not only do you pick up the sensory experience,
0:34:11 but you do pick up the changes that reflect the anxiety,
0:34:14 the helplessness, the other features
0:34:17 that constitute the entire picture
0:34:19 that a patient has.
0:34:20 – And to take it from the realm
0:34:23 of purely subjective to objective.
0:34:25 – Exactly, and that’s helping us
0:34:26 in our pre-clinical studies,
0:34:30 because again, pain is extremely complex,
0:34:31 there are many different features.
0:34:36 If we study those neurons that are the prime initiators
0:34:40 for pain and we can actually grow using stem cell technology,
0:34:42 we can take patient stem cells,
0:34:44 we can convert them into neurons,
0:34:47 we can look at features of the function of those neurons.
0:34:49 In the past, we used to look at one feature,
0:34:52 say the firing of action potentials
0:34:54 as a surrogate for their activity.
0:34:57 Now we recognize there are tens of features
0:35:00 of phenotypes as we call that constitute
0:35:02 the entire functional repertoire of these cells.
0:35:06 And again, machine learning is helping us see the patterns
0:35:09 and how they correlate with a normal state,
0:35:11 a disease state, what kind of disease state,
0:35:13 and the response to treatment.
0:35:17 So we now have tools at last to embrace complexity
0:35:21 and from that define what are the components
0:35:23 and which ones are relevant and important
0:35:25 and which ones to go for.
0:35:26 – And where to go next.
0:35:27 – And where to go next.
0:35:29 We are identifying other targets
0:35:30 that are looking very promising
0:35:35 and I’m optimistic that in consequence,
0:35:38 I think we are going to be able to make choices
0:35:40 about which are the best targets to go for,
0:35:43 find the means to identify the best modalities
0:35:46 to go for those targets,
0:35:49 to develop new ways of running clinical trials
0:35:50 that would be more sensitive,
0:35:52 both to suppress symptoms,
0:35:56 but also to prevent the evolution of permanent changes
0:35:57 in the function of the nervous system
0:36:00 that is a big driver of the chronicity of pain.
0:36:03 – Do you think that if we are able to get to that stage,
0:36:05 it would just roll through the system
0:36:08 and become adopted or would there be challenges?
0:36:10 – There are always challenges.
0:36:12 As is typical in biotech,
0:36:14 there are going to be many failures.
0:36:18 I think it is important that we go for risky
0:36:20 because some of them will pan out
0:36:24 but the portfolio of potential possibilities
0:36:26 is sufficiently exciting now
0:36:29 that I do think that we’ll be able to look back
0:36:31 in the near future and say,
0:36:34 at last we both understand pain
0:36:36 and we can now control it
0:36:39 in a way that has just simply not been possible before.
0:36:42 – I feel like we need a new thing in the doctor’s office
0:36:44 that says you’re experiencing one of these
0:36:48 four different types of pain for starters.
0:36:51 Thank you so much for joining us on the A16Z podcast.
0:36:51 – It’s a pleasure.
Imagina la audacia de las misiones Apolo: llegar a la luna no se logró lanzando cohetes al azar, sino diseñando meticulosamente cada componente y ensamblándolos paso a paso. Esta poderosa analogía describe el cambio sísmico que está ocurriendo en la biología, al pasar de una ciencia de alto riesgo y estocástica a una disciplina de ingeniería metódica y predecible. La conversación explora esta transición, donde el objetivo es identificar “Legos” biológicos fundamentales (partes estandarizadas e interoperables, como proteínas o circuitos genéticos) que puedan combinarse de forma confiable para construir terapias y sistemas complejos, de manera similar a como un ingeniero utiliza vigas para construir un puente.
Central en este cambio es la definición misma de un fármaco. Cuando las terapias eran moléculas simples, los “Legos” eran anillos químicos básicos. Ahora, a medida que los fármacos se convierten en entidades celulares y genéticas complejas, la necesidad de partes biológicas bien caracterizadas se vuelve crítica. Empresas como Asimov están liderando este espacio creando estos componentes fundamentales y las herramientas de diseño para ensamblarlos, buscando una alta previsibilidad. Esta mentalidad de ingeniería altera fundamentalmente la propuesta de valor en biotecnología. En lugar de que cada nuevo fármaco sea un proyecto científico personalizado y de alto riesgo, una plataforma de ingeniería aprende de cada iteración. Notablemente, esto significa que el segundo programa de fármacos puede ser más valioso que el primero, ya que el conocimiento se acumula, haciendo que el desarrollo sea más rápido, barato y efectivo con el tiempo.
Las implicaciones se extienden a todos los aspectos de la industria, desde el talento hasta los modelos de negocio. La academia responde con nuevos departamentos de bioingeniería, fusionando disciplinas como la ingeniería mecánica y la informática con la biología. El aprendizaje automático actúa como un potente acelerador, transformando el análisis de datos personalizado en un proceso de ingeniería repetible que puede descubrir patrones invisibles para los humanos. Para los emprendedores, esto cambia la estrategia de comercialización. El éxito ya no depende solo de un dramático momento “¡Eureka!” en un laboratorio, sino de demostrar un proceso predecible y escalable, demostrando que sabes dónde van los tornillos. Esta previsibilidad permite un modelo de negocio de “aterrizar y expandir”, antes poco común en biotecnología, donde una pequeña prueba de concepto se escala de manera confiable hacia asociaciones y portafolios más grandes.
Perspectivas Sorprendentes
- Los fracasos se vuelven fundamentales: En un paradigma de ingeniería, los falsos positivos y los experimentos fallidos no son callejones sin salida, sino puntos de datos cruciales para aprender y refinar el sistema, haciéndolos tan valiosos como los éxitos.
- La cronología no determina el valor: En la biotecnología tradicional, el fármaco candidato más avanzado es el más valioso. En la biología de ingeniería, el segundo o tercer activo suele ser más valioso que el primero, ya que la plataforma aprende y mejora con cada iteración.
- La plataforma es el producto: Históricamente, las plataformas terapéuticas estaban subvaloradas en favor de los activos farmacéuticos específicos que producían. Ahora, una plataforma capaz de generar múltiples activos de manera reproducible, predecible y basada en ingeniería se está convirtiendo en la ventaja competitiva central.
- El aprendizaje automático como fuerza democratizadora: El ML no se presenta como una IA mística, sino como la herramienta estadística definitiva para aprovechar los datos a gran escala. Puede transformar un proceso tradicionalmente dependiente de la intuición de un científico durante décadas en un flujo de trabajo de ingeniería sistemático y repetible.
Aplicaciones Prácticas
- Adopta una hoja de ruta de ingeniería: Divide una gran visión (por ejemplo, “curar el cáncer”) en una serie de hitos de ingeniería más pequeños y comprobables, similares a las etapas de los programas Mercury, Gemini y Apolo. Utiliza OKRs y KPIs para rastrear el progreso en estos pasos componentes.
- Busca previsibilidad sobre avances singulares: Al desarrollar una tecnología, concéntrate en demostrar un alto porcentaje de resultados predecibles y reproducibles (por ejemplo, “el 80% de nuestros circuitos genéticos diseñados funcionan como se espera”) en lugar de depender de resultados espectaculares y únicos.
- Diseña para la iteración y la acumulación de conocimiento: Construye sistemas y procesos que capturen datos de cada experimento, exitoso o no, para alimentar y mejorar el siguiente ciclo de diseño. Esta acumulación de conocimiento es el motor del progreso en ingeniería.
- Asóciate para demostrar escalabilidad: Para el desarrollo empresarial, estructura las alianzas iniciales como pruebas de concepto a corto plazo y bien delimitadas, diseñadas para demostrar tu valor predecible e iterativo. Esto genera la credibilidad necesaria para “aterrizar y expandir” con acuerdos más grandes.
En una fábrica de Mercedes en Alemania, los gerentes retiraron robots de automatización a gran escala y los reemplazaron con robots colaborativos que trabajan junto a empleados humanos, lo que generó mejoras drásticas en el rendimiento. Esta medida contraintuitiva—desautomatizar para añadir más personas—ilustra un tema central de la conversación: el futuro del trabajo no se trata de que las máquinas reemplacen a los humanos, sino de una asociación sofisticada en la que cada uno hace lo que mejor sabe hacer. Esta evolución está impulsada por un cambio de la producción en masa a la personalización masiva, donde los clientes esperan productos hechos a la medida de sus especificaciones exactas, lo que crea una necesidad de fabricación flexible y adaptable que los robots rígidos y anticuados jamás podrían manejar.
La discusión repasa la historia de la automatización laboral a través de tres generaciones distintas. La primera fue la era de la gestión científica de Henry Ford, incorporando el trabajo físico a los procesos industriales. La segunda, a finales de los 90, vio cómo los sistemas de TI automatizaban el trabajo del conocimiento pero resultaban en procesos estáticos e inflexibles. Ahora estamos entrando en una tercera generación definida por el trabajo dinámico, adaptable y personalizado. Aquí, la IA y el aprendizaje automático brindan orientación personalizada en tiempo real—como un gemelo digital de un motor de avión que le dice a un técnico el mantenimiento exacto necesario—y empoderan a los trabajadores para reconfigurar los cobots sobre la marcha sin necesidad de un programador. Esto transforma los empleos en lugar de eliminarlos, creando un rico “espacio intermedio faltante” de colaboración humano-máquina.
Dentro de este espacio intermedio faltante, están surgiendo dos nuevas familias de roles. Primero, empleos donde los humanos ayudan a las máquinas, como entrenadores de comportamiento de IA, explicadores y sostenes que aseguran que estos sistemas reflejen los valores de la empresa y operen de manera efectiva. Segundo, empleos donde las máquinas dan superpoderes a los humanos, como exoesqueletos para aumento físico o chatbots “compañeros de ala” que asisten a agentes de servicio al cliente manejando la búsqueda de información de fondo. La conclusión clave es que se están creando millones de estos nuevos roles, y los trabajos existentes, como el ingeniero industrial con cronómetro, están siendo potenciados con datos impulsados por IA, liberándolos de tareas de medición para enfocarse en interpretación de orden superior y gestión del cambio.
La conversación aborda directamente el miedo a la pérdida de empleos con matemáticas claras: con solo alrededor de 1,5 millones de robots en el planeta y 340 millones de personas en trabajos de producción, la automatización completa de todo el trabajo humano es un escenario lejano e irreal. El desafío y la oportunidad más inmediatos radican en construir organizaciones hábiles para aprovechar la IA. Esto requiere un cambio cultural hacia la experimentación continua, un enfoque obsesivo en la calidad de los datos y la inversión en habilidades. Es importante destacar que la IA más elegante a menudo se oculta a simple vista, integrada sin problemas como un corrector ortográfico, lo que impulsa la adopción y la confianza.
En última instancia, prepararse para este futuro tiene menos que ver con entrenar a las personas para pensar como máquinas y más con redoblar la apuesta en las fortalezas exclusivamente humanas. Habilidades como la improvisación, la empatía, la comunicación y la extrapolación (sacar conclusiones más allá de los datos) serán primordiales. Los trabajadores más efectivos serán aquellos que aprendan a interrogar inteligentemente los sistemas de IA, aplicando el juicio humano a los poderosos cálculos que proporcionan las máquinas.
### Perspectivas Sorprendentes
– Una fábrica de Mercedes logró mejoras significativas en el rendimiento al *desautomatizar*—reemplazando robots a gran escala con robots colaborativos (cobots) más pequeños y *aumentando* el número de trabajadores humanos en la línea.
– La demanda de personalización masiva es tan extrema que un vehículo popular como el Ford F-150 tiene más de *un billón* de combinaciones posibles de construcción, una complejidad manejable para humanos adaptables pero desalentadora para máquinas programadas tradicionalmente.
– A pesar de la ansiedad por la automatización, matemáticas simples muestran que el reemplazo total de empleos humanos en la fabricación está a generaciones de distancia: con 340 millones de trabajadores de producción y robots que reemplazan aproximadamente 5-6 trabajos cada uno, se necesitarían unos 60 millones de robots, mientras que la capacidad de producción global actual es de solo unos 250.000-500.000 por año.
– Una de las nuevas cadenas de herramientas más valiosas y sofisticadas para el desarrollo de software no se trata de escribir código, sino de crear complejos flujos de trabajo e interfaces para etiquetar datos—como el software complejo, similar a Photoshop, utilizado en Tesla para entrenar su IA.
– Un principio de diseño fundamental para una IA exitosa puede ser *ocultarla* al usuario final, integrándola tan sin problemas como un corrector ortográfico o una aplicación de mapas, lo que aumenta la confianza y la adopción al hacer que la tecnología se sienta natural e intuitiva.
### Conclusiones Prácticas
– **Enfócate en la Aumentación, No en el Reemplazo:** Al implementar la automatización, busca oportunidades en el “espacio intermedio faltante” donde humanos y máquinas colaboren, usando la IA para dar superpoderes a los empleados (por ejemplo, chatbots para agentes, información de datos para ingenieros) en lugar de buscar automatizar completamente sus roles.
– **Cultiva los Datos como un Hábito Central:** Para cualquier organización, la base del éxito de la IA es un enfoque obsesivo en la calidad de los datos y establecer una sólida “cadena de suministro de datos”. Prioriza la medición y trata los datos como un activo crítico.
– **Construye para la Experimentación:** Adopta una mentalidad iterativa y experimental similar a las pruebas A/B en la web. Diseña sistemas y procesos que permitan pruebas rápidas, aprendizaje del fracaso y adaptación continua de las aplicaciones de IA.
– **Invierte en el Desarrollo de Habilidades Centradas en el Humano:** Para tu fuerza laboral, invierte menos en entrenar a las personas para pensar como máquinas y más en habilidades donde los humanos sobresalen: improvisación, comunicación, juicio ético y la capacidad de extrapolar y hacer las preguntas correctas a un sistema de IA.
– **Mapea el Problema con la Técnica Correcta:** Al aplicar la IA, dedica un tiempo significativo a definir claramente el problema.
El diferenciador clave es elegir la técnica de IA adecuada para el desafío específico, no simplemente aplicar IA por sí misma.
Imagine a audácia das missões Apollo: ir à Lua não foi conseguido testando foguetes aleatoriamente, mas sim através da engenharia meticulosa de cada componente e sua montagem passo a passo. Esta é a poderosa analogia usada para descrever a mudança sísmica que está a ocorrer na biologia, à medida que esta transita de uma ciência de alto risco e estocástica para uma disciplina de engenharia metódica e previsível. A conversa explora esta transição, cujo objetivo é identificar os “Legos” biológicos fundamentais — peças padronizadas e interoperáveis, como proteínas ou circuitos genéticos — que possam ser misturadas e combinadas de forma confiável para construir terapias e sistemas complexos, tal como um engenheiro usa vigas para construir uma ponte.
Central a esta mudança está a própria definição de um fármaco. Quando as terapias eram moléculas simples, os “Legos” eram anéis químicos básicos. Agora, à medida que os fármacos se tornam entidades celulares e genéticas complexas, a necessidade de partes biológicas bem caracterizadas torna-se crítica. Empresas como a Asimov estão a pioneirar neste espaço, criando estes componentes fundamentais e as ferramentas de design para os montar, visando uma alta previsibilidade. Esta mentalidade de engenharia altera fundamentalmente a proposição de valor na biotecnologia. Em vez de cada novo fármaco ser um projeto científico personalizado e de alto risco, uma plataforma de engenharia aprende com cada iteração. Surpreendentemente, isto significa que o segundo programa de fármacos pode ser mais valioso do que o primeiro, à medida que o conhecimento se acumula, tornando o desenvolvimento mais rápido, barato e eficaz ao longo do tempo.
As implicações repercutem-se em todos os aspetos da indústria, desde o talento até aos modelos de negócio. A academia está a responder com novos departamentos de bioengenharia, fundindo disciplinas como a engenharia mecânica e a ciência da computação com a biologia. A aprendizagem automática atua como um acelerador poderoso, transformando a análise de dados personalizada num processo de engenharia repetível que pode revelar padrões invisíveis aos humanos. Para os empreendedores, isto muda a estratégia de entrada no mercado. O sucesso já não depende apenas de um dramático momento “Eureka!” num laboratório, mas de demonstrar um processo previsível e escalável — provando que se sabe onde colocar os parafusos. Esta previsibilidade permite um modelo de negócio “aterrar e expandir”, anteriormente raro na biotecnologia, em que uma pequena prova de conceito escala de forma confiável para parcerias e portfólios maiores.
Em uma fábrica da Mercedes na Alemanha, os gestores retiraram robôs automatizados de grande escala e os substituíram por robôs colaborativos que trabalham junto com funcionários humanos, resultando em ganhos drásticos de desempenho. Esse movimento contraintuitivo — desautomatizar para incluir mais pessoas — ilustra um tema central da discussão: o futuro do trabalho não se trata de máquinas substituindo humanos, mas sim de uma parceria sofisticada na qual cada um faz o que faz de melhor. Essa evolução está sendo impulsionada pela transição da produção em massa para a personalização em massa, em que os clientes esperam produtos sob medida para suas especificações exatas, criando a necessidade de uma manufatura flexível e adaptativa que os robôs antigos e rígidos jamais poderiam lidar.
A discussão traça a história da automação do trabalho por meio de três gerações distintas. A primeira foi a era da gestão científica de Henry Ford, integrando trabalho físico aos processos industriais. A segunda, no fim dos anos 90, viu sistemas de TI automatizarem o trabalho intelectual, mas resultou em processos estáticos e inflexíveis. Agora estamos entrando em uma terceira geração, definida pelo trabalho dinâmico, adaptativo e personalizado. Nela, a inteligência artificial e o aprendizado de máquina oferecem orientação personalizada em tempo real — como um “gêmeo digital” de um motor de jato informando a um técnico a manutenção exata necessária — e capacitam os trabalhadores a reconfigurar cobots em tempo real sem a necessidade de um programador. Isso transforma os empregos em vez de eliminá-los, criando um rico “meio-termo ausente” de colaboração homem-máquina.
Dentro desse meio-termo ausente, estão surgindo duas novas famílias de funções. Primeiro, empregos em que humanos ajudam máquinas, como treinadores comportamentais de IA, explicadores e mantenedores que garantem que esses sistemas reflitam os valores da empresa e operem com eficácia. Segundo, empregos em que máquinas dão superpoderes aos humanos, como exoesqueletos para ampliação física ou chatbots “colegas de asa” que auxiliam agentes de atendimento ao cliente gerenciando consultas de informações em segundo plano. A conclusão-chave é que milhões dessas novas funções estão sendo criadas, e empregos existentes, como o do engenheiro industrial com um cronômetro, estão sendo potencializados com dados orientados por IA, libertando-os de tarefas de medição para focarem em interpretação de ordem superior e gestão de mudanças.
A discussão aborda diretamente o medo da perda de empregos com matemática clara: com apenas cerca de 1,5 milhão de robôs no planeta e 340 milhões de pessoas em empregos de produção, a automação completa de todo o trabalho humano é um cenário distante e irrealista. O desafio e a oportunidade mais imediatos residem na construção de organizações hábeis em aproveitar a IA. Isso requer uma mudança cultural em direção à experimentação contínua, um foco obsessivo na qualidade dos dados e investimento em habilidades. Importante destacar que a IA mais elegante muitas vezes se esconde à vista de todos, integrada de forma tão harmoniosa quanto um corretor ortográfico, o que impulsiona sua adoção e confiança.
Por fim, preparar-se para esse futuro tem menos a ver com treinar pessoas para pensarem como máquinas e mais com reforçar as forças exclusivamente humanas. Habilidades como improvisação, empatia, comunicação e extrapolação (tirar conclusões além dos dados) serão primordiais. Os trabalhadores mais eficazes serão aqueles que aprenderem a interrogar de forma inteligente os sistemas de IA, aplicando o julgamento humano aos poderosos cálculos fornecidos pelas máquinas.
### **Perspectivas Surpreendentes**
– Uma fábrica da Mercedes alcançou melhorias significativas de desempenho ao *desautomatizar* — substituindo robôs de grande porte por robôs colaborativos menores (cobots) e *aumentando* o número de trabalhadores humanos na linha.
– A busca pela personalização em massa é tão intensa que um veículo popular como o Ford F-150 tem mais de *um trilhão* de combinações de configuração possíveis, uma complexidade gerenciável por humanos adaptáveis, mas assustadora para máquinas programadas tradicionalmente.
– Apesar da ansiedade em relação à automação, uma matemática simples mostra que a substituição total de empregos humanos na manufatura está a gerações de distância: com 340 milhões de trabalhadores na produção e estimando que cada robô substitua de 5 a 6 empregos, seriam necessários cerca de 60 milhões de robôs, enquanto a capacidade de produção global atual é de apenas cerca de 250.000 a 500.000 por ano.
– Uma das cadeias de ferramentas mais valiosas e sofisticadas para o desenvolvimento de software não se trata de escrever código, mas sim de criar fluxos e interfaces complexas para rotulação de dados — como o software complexo, semelhante ao Photoshop, usado na Tesla para treinar sua IA.
– Um princípio de design central para uma IA bem-sucedida pode ser *ocultá-la* do usuário final, integrando-a de forma tão harmoniosa quanto um corretor ortográfico ou um aplicativo de mapas, o que aumenta a confiança e a adoção ao fazer a tecnologia parecer natural e intuitiva.
### **Ações Práticas**
– **Foque em Aumentar, Não Substituir:** Ao implementar automação, busque oportunidades no “meio-termo ausente”, onde humanos e máquinas colaboram, usando a IA para dar superpoderes aos funcionários (por exemplo, chatbots para agentes, insights de dados para engenheiros) em vez de buscar automatizar totalmente suas funções.
– **Cultive o Dado como um Hábito Central:** Para qualquer organização, a base do sucesso da IA é um foco obsessivo na qualidade dos dados e no estabelecimento de uma “cadeia de suprimentos de dados” robusta. Priorize a medição e trate os dados como um ativo crítico.
– **Projete para Experimentação:** Adote uma mentalidade iterativa e experimental, semelhante aos testes A/B na web. Crie sistemas e processos que permitam testes rápidos, aprendizado com os erros e adaptação contínua das aplicações de IA.
– **Invista no Desenvolvimento de Habilidades Centradas no Ser Humano:** Para sua força de trabalho, invista menos em treinar as pessoas para pensarem como máquinas e mais em habilidades em que os humanos se destacam: improvisação, comunicação, julgamento ético e a capacidade de extrapolar e fazer as perguntas certas a um sistema de IA.
– **Mapeie o Problema para a Técnica Correta:** Ao aplicar IA, dedique um tempo significativo para definir claramente o problema.
O diferencial principal é escolher a técnica de IA apropriada para o desafio específico, não apenas aplicar IA por aplicá-la.
Central a esta mudança está a própria definição de um fármaco. Quando as terapias eram moléculas simples, os “Legos” eram anéis químicos básicos. Agora, à medida que os fármacos se tornam entidades celulares e genéticas complexas, a necessidade de partes biológicas bem caracterizadas torna-se crítica. Empresas como a Asimov estão a pioneirar neste espaço, criando estes componentes fundamentais e as ferramentas de design para os montar, visando uma alta previsibilidade. Esta mentalidade de engenharia altera fundamentalmente a proposição de valor na biotecnologia. Em vez de cada novo fármaco ser um projeto científico personalizado e de alto risco, uma plataforma de engenharia aprende com cada iteração. Surpreendentemente, isto significa que o segundo programa de fármacos pode ser mais valioso do que o primeiro, à medida que o conhecimento se acumula, tornando o desenvolvimento mais rápido, barato e eficaz ao longo do tempo.
As implicações repercutem-se em todos os aspetos da indústria, desde o talento até aos modelos de negócio. A academia está a responder com novos departamentos de bioengenharia, fundindo disciplinas como a engenharia mecânica e a ciência da computação com a biologia. A aprendizagem automática atua como um acelerador poderoso, transformando a análise de dados personalizada num processo de engenharia repetível que pode revelar padrões invisíveis aos humanos. Para os empreendedores, isto muda a estratégia de entrada no mercado. O sucesso já não depende apenas de um dramático momento “Eureka!” num laboratório, mas de demonstrar um processo previsível e escalável — provando que se sabe onde colocar os parafusos. Esta previsibilidade permite um modelo de negócio “aterrar e expandir”, anteriormente raro na biotecnologia, em que uma pequena prova de conceito escala de forma confiável para parcerias e portfólios maiores.
Ideias Surpreendentes
- Os fracassos tornam-se fundamentais: Num paradigma de engenharia, falsos positivos e experiências falhadas não são becos sem saída, mas pontos de dados cruciais para aprender e refinar o sistema, tornando-os tão valiosos quanto os sucessos.
- A cronologia não determina o valor: Na biotecnologia tradicional, o candidato a fármaco mais avançado é o mais valioso. Na biologia de engenharia, o segundo ou terceiro ativo é frequentemente mais valioso do que o primeiro, à medida que a plataforma aprende e melhora com cada iteração.
- A plataforma é o produto: Historicamente, as plataformas terapêuticas eram subvalorizadas em favor dos ativos farmacêuticos específicos que produziam. Agora, uma plataforma capaz de gerar múltiplos ativos de forma reprodutível, previsível e orientada pela engenharia está a tornar-se a vantagem competitiva central.
- A aprendizagem automática como força democratizante: A AM é enquadrada não como uma IA mística, mas como a ferramenta estatística definitiva para aproveitar dados em larga escala. Pode transformar um processo tradicionalmente dependente da intuição de um cientista ao longo de décadas num fluxo de trabalho de engenharia sistemático e repetível.
Aplicações Práticas
- Adotar um roteiro de engenharia: Divida uma grande visão (por exemplo, “curar o cancro”) numa série de marcos de engenharia menores e testáveis, semelhantes às fases dos programas Mercury, Gemini e Apollo. Utilize OKRs e KPIs para monitorizar o progresso nestes passos componentes.
- Priorizar a previsibilidade em vez de descobertas isoladas: Ao desenvolver uma tecnologia, foque-se em demonstrar uma alta percentagem de resultados previsíveis e reprodutíveis (por exemplo, “80% dos nossos circuitos genéticos projetados funcionam conforme o previsto”) em vez de depender de resultados sensacionais únicos.
- Projetar para iteração e acumulação de conhecimento: Construa sistemas e processos que capturem dados de cada experiência, bem-sucedida ou não, para os reintegrar e melhorar o próximo ciclo de design. Este conhecimento acumulado é o motor do progresso da engenharia.
- Parcerias para provar a escalabilidade: Para o desenvolvimento de negócios, estruture parcerias iniciais como provas de conceito de curto prazo e bem delimitadas, concebidas para demonstrar o seu valor previsível e iterativo. Isto constrói a credibilidade necessária para “aterrar e expandir” com acordos maiores.
Em uma fábrica da Mercedes na Alemanha, os gestores retiraram robôs automatizados de grande escala e os substituíram por robôs colaborativos que trabalham junto com funcionários humanos, resultando em ganhos drásticos de desempenho. Esse movimento contraintuitivo — desautomatizar para incluir mais pessoas — ilustra um tema central da discussão: o futuro do trabalho não se trata de máquinas substituindo humanos, mas sim de uma parceria sofisticada na qual cada um faz o que faz de melhor. Essa evolução está sendo impulsionada pela transição da produção em massa para a personalização em massa, em que os clientes esperam produtos sob medida para suas especificações exatas, criando a necessidade de uma manufatura flexível e adaptativa que os robôs antigos e rígidos jamais poderiam lidar.
A discussão traça a história da automação do trabalho por meio de três gerações distintas. A primeira foi a era da gestão científica de Henry Ford, integrando trabalho físico aos processos industriais. A segunda, no fim dos anos 90, viu sistemas de TI automatizarem o trabalho intelectual, mas resultou em processos estáticos e inflexíveis. Agora estamos entrando em uma terceira geração, definida pelo trabalho dinâmico, adaptativo e personalizado. Nela, a inteligência artificial e o aprendizado de máquina oferecem orientação personalizada em tempo real — como um “gêmeo digital” de um motor de jato informando a um técnico a manutenção exata necessária — e capacitam os trabalhadores a reconfigurar cobots em tempo real sem a necessidade de um programador. Isso transforma os empregos em vez de eliminá-los, criando um rico “meio-termo ausente” de colaboração homem-máquina.
Dentro desse meio-termo ausente, estão surgindo duas novas famílias de funções. Primeiro, empregos em que humanos ajudam máquinas, como treinadores comportamentais de IA, explicadores e mantenedores que garantem que esses sistemas reflitam os valores da empresa e operem com eficácia. Segundo, empregos em que máquinas dão superpoderes aos humanos, como exoesqueletos para ampliação física ou chatbots “colegas de asa” que auxiliam agentes de atendimento ao cliente gerenciando consultas de informações em segundo plano. A conclusão-chave é que milhões dessas novas funções estão sendo criadas, e empregos existentes, como o do engenheiro industrial com um cronômetro, estão sendo potencializados com dados orientados por IA, libertando-os de tarefas de medição para focarem em interpretação de ordem superior e gestão de mudanças.
A discussão aborda diretamente o medo da perda de empregos com matemática clara: com apenas cerca de 1,5 milhão de robôs no planeta e 340 milhões de pessoas em empregos de produção, a automação completa de todo o trabalho humano é um cenário distante e irrealista. O desafio e a oportunidade mais imediatos residem na construção de organizações hábeis em aproveitar a IA. Isso requer uma mudança cultural em direção à experimentação contínua, um foco obsessivo na qualidade dos dados e investimento em habilidades. Importante destacar que a IA mais elegante muitas vezes se esconde à vista de todos, integrada de forma tão harmoniosa quanto um corretor ortográfico, o que impulsiona sua adoção e confiança.
Por fim, preparar-se para esse futuro tem menos a ver com treinar pessoas para pensarem como máquinas e mais com reforçar as forças exclusivamente humanas. Habilidades como improvisação, empatia, comunicação e extrapolação (tirar conclusões além dos dados) serão primordiais. Os trabalhadores mais eficazes serão aqueles que aprenderem a interrogar de forma inteligente os sistemas de IA, aplicando o julgamento humano aos poderosos cálculos fornecidos pelas máquinas.
### **Perspectivas Surpreendentes**
– Uma fábrica da Mercedes alcançou melhorias significativas de desempenho ao *desautomatizar* — substituindo robôs de grande porte por robôs colaborativos menores (cobots) e *aumentando* o número de trabalhadores humanos na linha.
– A busca pela personalização em massa é tão intensa que um veículo popular como o Ford F-150 tem mais de *um trilhão* de combinações de configuração possíveis, uma complexidade gerenciável por humanos adaptáveis, mas assustadora para máquinas programadas tradicionalmente.
– Apesar da ansiedade em relação à automação, uma matemática simples mostra que a substituição total de empregos humanos na manufatura está a gerações de distância: com 340 milhões de trabalhadores na produção e estimando que cada robô substitua de 5 a 6 empregos, seriam necessários cerca de 60 milhões de robôs, enquanto a capacidade de produção global atual é de apenas cerca de 250.000 a 500.000 por ano.
– Uma das cadeias de ferramentas mais valiosas e sofisticadas para o desenvolvimento de software não se trata de escrever código, mas sim de criar fluxos e interfaces complexas para rotulação de dados — como o software complexo, semelhante ao Photoshop, usado na Tesla para treinar sua IA.
– Um princípio de design central para uma IA bem-sucedida pode ser *ocultá-la* do usuário final, integrando-a de forma tão harmoniosa quanto um corretor ortográfico ou um aplicativo de mapas, o que aumenta a confiança e a adoção ao fazer a tecnologia parecer natural e intuitiva.
### **Ações Práticas**
– **Foque em Aumentar, Não Substituir:** Ao implementar automação, busque oportunidades no “meio-termo ausente”, onde humanos e máquinas colaboram, usando a IA para dar superpoderes aos funcionários (por exemplo, chatbots para agentes, insights de dados para engenheiros) em vez de buscar automatizar totalmente suas funções.
– **Cultive o Dado como um Hábito Central:** Para qualquer organização, a base do sucesso da IA é um foco obsessivo na qualidade dos dados e no estabelecimento de uma “cadeia de suprimentos de dados” robusta. Priorize a medição e trate os dados como um ativo crítico.
– **Projete para Experimentação:** Adote uma mentalidade iterativa e experimental, semelhante aos testes A/B na web. Crie sistemas e processos que permitam testes rápidos, aprendizado com os erros e adaptação contínua das aplicações de IA.
– **Invista no Desenvolvimento de Habilidades Centradas no Ser Humano:** Para sua força de trabalho, invista menos em treinar as pessoas para pensarem como máquinas e mais em habilidades em que os humanos se destacam: improvisação, comunicação, julgamento ético e a capacidade de extrapolar e fazer as perguntas certas a um sistema de IA.
– **Mapeie o Problema para a Técnica Correta:** Ao aplicar IA, dedique um tempo significativo para definir claramente o problema.
O diferencial principal é escolher a técnica de IA apropriada para o desafio específico, não apenas aplicar IA por aplicá-la.
What is the nature of physical pain? Why do we even experience it? Is there one type, or many? Do people experience pain differently? What is happening in our brains and our bodies when we experience pain? What is the biological link between pain and addiction? In this episode Clifford Woolf, Professor of Neurobiology at Harvard Medical School and a renowned expert on understanding pain, shares with with a16z’s Hanne Tidnam all we know about the biology of pain.
Technology is enabling a new, deeper, and much more complex understanding of pain—which pathways and neurons are activated in the brain when, what patterns might represent which experiences of pain. We now understand that the notion of pain as a simple switch that can be switched on or off (you have pain/you don’t have pain) and measured by categories like mild, moderate, or severe is just incorrect. Woolf describes the 4 different broad types of pain we in fact experience, what the purpose of each is, and what it means now that we can phenotype them and begin to understand them as distinct. Now that we have this deeper and much more complex understanding of pain, what does it mean for how we can treat pain in the future, and where we can intervene?
Leave a Reply
You must be logged in to post a comment.