SRF Interviews James Gold—Posted 16 August 2013
Interviewed by Hakon Heimer and Michele Solis
Psychiatry) (see SRF related news story), as well as on a new rating scale called the Clinical Assessment Interview for Negative Symptoms (CAINS) that was described in an American Journal of Psychiatry paper this year (see SRF related news story).
Negative symptoms have remained a backwater of study for the schizophrenia research field, with no clear direction given by anatomy, physiology, pharmacology, or psychology. However, a few intrepid researchers have ventured into these waters, and in the past year and half they have published some research results. Last year, SRF reported on two experimental studies that appeared in the Archives of General Psychiatry (now JAMA
SRF spoke several times with advisor Jim Gold of the Maryland Psychiatric Research Center about the puzzling nature of negative symptoms, and here we present excerpts from those conversations for the benefit of SRF readers.
SRF: Negative symptomatology is less familiar and seems more complicated for the non-clinician reader. What are negative symptoms, and are they inherently more complicated or less precise than positive symptoms?
Gold: Negative symptoms in general are defined as the absence of normal function. So, for example, a very common symptom, although it's probably not one of the more important, is blunted affect. When you conduct an interview to rate negative symptoms, you basically make a judgment about how decreased patients' facial mobility is, how much their expressive gesture is decreased. It can be done very reliably with training, but it's different from asking them, “Do you hear voices?” If they say they are hearing voices throughout the day, you have an idea of what it is that they are experiencing. It's very different when you are making judgments about how interested people are in goal-directed activity, pursuing pleasurable activities, socializing with other people, how much they enjoy things when they do them. So I do think that it's a more complex enterprise to evaluate the absence of things rather than the abnormal presence of things.
SRF: How do negative symptoms interact with medication? Do you have to be able to see patients unmedicated to really say these people have profound negative symptoms?
Gold: Obviously, there is a concern that conventional antipsychotics cause Parkinsonian symptoms, including reduced facial mobility and changes in vocal inflection. So it certainly is the case that high doses of medicine can create something that looks a lot like negative symptoms. That said, these things were really well described by Kraepelin in unmedicated patients. The Kraepelin textbook is amazing on this stuff because you feel like you’re reading about the patients that you are seeing today. When I worked at St. Elizabeth's with people who had supervised drug withdrawal periods, there were very few people whose negative symptoms improved, maybe a handful. And those were in the days of higher doses of medicine than we typically use now. So I don't think there is a lot of danger that negative symptoms are an artifact of antipsychotic treatment.
I can imagine that there are some people in whom it's clearly complicating. For example, there is an older study from Shitij Kapur showing increased levels of some type of dysphoria with higher levels of D2 receptor blockade [Mizrahi et al., 2007]. Now, dysphoria is different from negative symptoms, but I imagine that there are some people for whom the subjective experience of taking antipsychotics is somewhat unpleasant and demotivating. I hope and I believe that that's not what we are studying most of the time.
SRF: How are negative symptoms fundamentally different from depression, and is there an overlap of some sort?
Gold: Brian Kirkpatrick did some nice studies some years back on the deficit syndrome. The deficit syndrome in patients is defined as severe negative symptoms that are judged to be unrelated to medication side effects and unrelated to depression. When you look at depression ratings in those people, say on the BPRS [Brief Psychiatric Rating Scale], deficit syndrome patients were less depressed than non-deficit patients, and they were less likely to commit suicide in the Chestnut Lodge follow-up studies. So this is evidence that, in general, people who are troubled by negative symptoms are not also severely dysphoric and depressed. I think there is some recent evidence—it hasn't fully seen the light of day yet—that people with negative symptoms may have higher levels of trait negative affectivity. They might be somewhat distressed on some level much of the time, but not in an episode of depression.
SRF: Is that your data or someone else's?
Gold: Greg Strauss and I had a recent paper about emotion-attention interactions showing an effect of negative symptoms on how sensitive people were to negative stimuli [Strauss et al., 2011]. There are also data from, I believe, Alex Cohen on laboratory measures of emotional experience in people with schizophrenia, showing that if you ask people to make both a pleasantness and an unpleasantness rating on a particular experience or event that, oddly, when patients with schizophrenia say something is mostly pleasant, they will also endorse the fact that it is somewhat mildly unpleasant—as if there are a lot of affective blends going on [Cohen and Minor, 2010]. I suspect that that may be true of our negative symptom folks—that the impact of a positive emotion is mitigated by the presence of an ongoing negative affective trait.
Greg and I have yet to try to fully reconcile the line of work that he is really leading on emotion and emotional experience with the reinforcement learning approach that I've taken. We need to see some more of his results and do some of the studies in the exact same people at the same time and see how things relate to one another, and we haven't done that yet.
SRF: Maybe this is putting you on the spot since you are still thinking about this in different ways, but do you have a fundamental theory of negative symptoms, and how would it be similar to or different from other theories?
Gold: There are not really theories of negative symptoms; there are just descriptions of them. But one idea—I suppose you could call it a theory—is that a negative symptom such as reduced motivation is caused by a primary failure of hedonic experience. So, if you don't like anything, why would you want it? That would be the most straightforward, common sense explanation for lack of motivated behavior. The problem about that theory is there is a lot of evidence that in the moment, patients, including patients with negative symptoms, say that they do like things. Patients say that they like things about the same amount as healthy people do.
What may differ, though, is the brain signal that goes with that. For example, Jim Waltz had a paper in Neuropsychopharmacology a couple of years ago [Waltz et al., 2009] where we did a conditioning experiment in which you saw a light come on and then six seconds later you would get a small squirt of juice while you were in the scanner, and you were thirsty. Our patient participants said that they enjoyed and liked the juice just as much as healthy folks did. But the sort of brain signal that we saw in healthy volunteers that related to tasting the juice was altered in people with schizophrenia.
In contrast, there were some trials where you saw the light but you didn't get a squirt of juice, and that is thought to be a negative prediction error from a reinforcement learning point of view. Patients had the exact same BOLD signal decrease in reward-related areas as did healthy controls when they were disappointed by an outcome. But they didn't have the same brain response when they had a surprising positive outcome, when the juice came on at an odd, unexpected time. Their neural activity associated with receiving an unexpected reward was different from that of healthy controls, but their actual subjective report of their experience was the same. There have been other studies that have found similar sorts of things, so it could be that the subjective report of patients is far more normative than is the neural response to positive outcomes. It could be that the thing that's important to drive learning and behavior is the neural response rather than the subjective report.
But in terms of clear, convincing evidence that the thing that is driving reductions in motivated behavior is a primary alteration in hedonic experience, I think most people would say that is lacking in the literature.
As for our theory, we wrote a paper a number of years ago reviewing a bunch of our behavior experiments and some electrophysiological experiments, and we came to the broader conclusion that schizophrenia patients with negative symptoms have trouble representing the relative value of rewards and therefore might show alterations in decision making [Gold et al., 2008]. That was some of the background that led to the current work.
This is related to an influential view in the field pioneered by Ann Kring on consummatory and anticipatory pleasure. Ann's idea is that patients actually enjoy the experience of a reward as much as anyone else, but they don't show as much anticipatory pleasure about it, and that would de-hitch it from driving behavior [Chan et al., 2010]. In some ways I think the work we are doing is a more formal, learning-based approach to a related idea.
SRF: In your current work, are you taking on a subset of negative symptoms, and if so, what would you be including or excluding?
Gold: I'm most interested in avolition and amotivation. There have been a number of factor analyses of negative symptoms. The expressive, so to speak, negative symptoms—reduced gesture, reduced facial expression, reduced verbal output—tend to define one factor. The more behavioral symptoms—the lack of goal-directed behavior, lack of social behavior, lack of apparent enjoyment of potentially rewarding activities—tend to be a second factor. I'm much more interested in behavior than I am in the expressive stuff because I think the behavioral dimension is directly related to the severity of disability. In most of our studies we find much more robust relationships between experimental results and the behavioral ratings than with the expressivity ratings, which is just fine with me because I'm actually much more interested in why patients have so much trouble doing stuff and pursuing stuff rather than whether they smile about it when they do it.
SRF: What did you investigate in your 2012 Archives of General Psychiatry study [see SRF related news story]?
Gold: We had people learn, from feedback, four different-stimulus response associations. In two of the cases, if they chose the right response, they got a positive outcome—they got actual money; the wrong response got them nothing. For another two stimuli, if you learned the right response, the best you could do was to avoid losing money, to keep the money you had already earned; the wrong response took money away—a negative outcome. I think the critical thing that we found was that people with high levels of negative symptoms are actually better able to learn how to avoid losses than they are at learning how to achieve gains.
If our results are reproducible and reliable—and I believe we have shown that we've replicated them—they are a behavioral recipe for amotivation. If you are better able to learn what not to do than what to do, the most likely outcome is that you won't do a lot. That's the phenomenon that we are trying to explain: Why do patients have a reduced behavioral repertoire? Why do they seem to have as much trouble as they do sustaining and engaging in goal-directed behavior and pursuing long-term goals?
Our belief is that we all experience a mix of positive and negative outcomes. We think that people with negative symptoms are in essence more tuned to learn from negative outcomes than they are from positive outcomes. If that's true, it's totally cool because it actually makes sense.
SRF: Has this been studied in normal populations, where there presumably is a spectrum of people who do one or the other better, or do both very well?
Gold: It's been studied some. In general, people are better at learning from positive outcomes than negative outcomes. That was true in our healthy control data, where the learning rates and final asymptotic levels of learning were higher for the positive outcomes than the negative outcomes. In part, you would expect that from the simple stimulus-response association point of view. When you get a positive outcome, you know that you did the right thing. When you experience a negative outcome, you know you did the wrong thing, but you don't necessarily learn simultaneously that the alternative is better. So I think that the learning of certain positively reinforced responses is a little bit more direct than is the learning to make an alternative response to avoid a negative outcome.
The thing that was interesting about our data was that they best fit with a computational model that combined an “actor-critic” model that's driven purely by prediction-error learning with a “Q-learning” module that weighs the value of the choices it's making at the time of decision making. What was remarkable about the way the data turned out was that if we just took out this extra Q-learning module, the actor-critic model fit the negative symptom patient data very, very well. In order to fit the healthy control data, though, we needed to add this other Q-learning influence, which is thought to be a top-down, orbitofrontal cortex role in representing the relative value of choices.
After we had patients learn in the initial training phase, we did what's called a transfer phase. We took all of the stimulus pairs that they had originally been presented with, where one was better than the other, and we now mixed up all the stimuli. People then had to express their preferences out of the new pairings. They got no further feedback. For example, in one transfer pair, one stimulus was a 90 percent gain stimulus and the other was a 90 percent successful loss-avoidance stimulus, and we paired those together. Healthy controls, and patients with schizophrenia with low negative symptoms, and this Q-value model plus actor-critic model preferred the stimulus that had been associated with positive outcomes—the actual reward receipt. The high negative symptom patients were indifferent to whether one was associated with reward or punishment avoidance. A pure actor-critic model that only learns by prediction errors shows no preference either, because as the actor-critic is going along learning the stimuli, the 90 percent win stimulus and the 90 percent loss-of-win stimulus are the best alternative that's available in the trial. For a pure actor-critic model, those stimuli are of equal value at the end of the day, and that's how our high negative symptom patients behaved.
The important thing about that from a larger theoretical neurobiological point of view is that one way of understanding negative symptoms—from a reinforcement learning perspective—would be that these folks don't signal positive prediction errors. But by showing that we can fit the patient data with the actor-critic model, we are able to show that they are learning from positive prediction errors. They just are not learning the difference in value between a positive prediction error that's associated with an actual positive outcome as opposed to a positive prediction error that is associated with successful loss avoidance.
SRF: So they've lost the Q-learning input?
Gold: Yes. From our point of view, they've lost the ability to represent the potential reward value of choices and outcomes in a way that can reliably guide behavior. They have, unfortunately, maintained the ability to code negative prediction errors when things are worse than expected and learn to make behavioral adaptations to avoid things that are associated with those outcomes.
To go back to where we started, the reason we were really excited is that this is a coherent and mechanistic view of how to produce an absence of motivated behavior. It's not just saying that patients don't do anything because they have bad brains. It's not just saying patients don't do things because they don't enjoy things, because we know that is not true. It's coming up with a psychologically plausible, computationally mechanistic model of how you would build a reinforcement learning system that results in the tragedy of negative symptoms.
SRF: So where do you go from here, say, from the perspective of cementing the psychology findings, or in terms of trying to connect it to neurobiology?
Gold: [laughing] Well, I thought we had some great ideas, but our grant reviewers didn't agree. Actually, the thing we'd like to do is try to get direct neural evidence for this. We have some ideas on how to do that by recording EEG activity while people do reinforcement learning tasks and then using computational modeling of the EEG data to test whether the abnormality we think happens occurs at the time of decision as opposed to the time of the outcome. That would help nail down this idea that people have trouble representing prospective value rather than processing outcomes.
SRF: What are your thoughts about the approach taken by Michael Green and colleagues to studying negative symptoms [see SRF related news story], which seems very different from yours?
Gold: I don’t do structural equation modeling, and I’m very impressed at how you can take a table of intercorrelations like the one that they’ve published and turn it into this subtly elegant thing. I wish I knew how to do it. The path that they test—going from early visual perception to social cognition to defeatist beliefs to negative symptoms to functional outcomes—that’s a very nice story. And I actually don’t think it’s terribly inconsistent with things that I believe. We believe that motivational problems result, in part at least, from people having difficulty representing the value of positive outcomes and then activating behavior to achieve those outcomes. The Beck model of defeatist beliefs is sort of consistent with that. They have a scale that measures low expectations of success and of pleasure in social interactions. I don’t know that our ideas are actually that different. They’re posed at very different levels of description, but I don’t think that they’re different.
The Beck group—principally Aaron Beck, and also Paul Grant has done a lot of this work with him—believes that negative life outcomes that come from having cognitive impairments or social cognitive impairments lead to a set of beliefs about being inadequate and that it’s not worth trying—things are going to turn out badly—and that the more you believe those sorts of things, it makes sense that you’re not terribly motivated to try and do new things or push very hard at things that aren’t going well. From our framework, patients have learned from negative outcomes, and that will look like something that clinicians will rate as avolition and anhedonia. That also leads to poor ratings of functional outcome, which are really almost tautological at that point, because it’s not really saying that negative symptoms cause poor functional outcomes, but the things that you’re assessing as negative symptoms are very similar to the things you’re assessing as outcome: Are people working? How rich is their social network? Do they have interests? Do they have hobbies? So it's funny that in the literature, people often will talk about negative symptoms predicting outcomes. Negative symptoms are outcome, for the ones that have to do with action. Flat affect is not outcome; reduced verbal output is not outcome.
SRF: Can you describe the main goals of your article with Greg Strauss [see SRF related news story] on anhedonia, which seems to be in the realm of general rather than social cognition?
Gold: The Strauss paper is addressing a narrower issue. Is there something about the way people report on their past experiences that then project into their anticipation of future emotional experiences? If you ask patients how much they enjoyed doing things in the past or how much they anticipate doing things in the future, they tend to respond in ways that are different from controls by reporting less pleasure. But if you ask them in the moment—"How much do you like this movie?" or "How much do you enjoy eating that dessert?"—their answers tend to be very similar to those of controls. Greg was trying to address that paradox: Is there something about the way these questions are asked and the way people go about answering them that might explain the apparent contradictions in the literature between reports of past versus in-the-moment pleasure and in-the-moment pleasure with anticipated future pleasure? Greg’s focus was on the report of emotional experience, which overlaps with the rating of anhedonia on the SANS [Scale for the Assessment of Negative Symptoms] but does not address the other negative symptoms that get rated on that scale.
SRF: So you would fit the Strauss paper in a narrower portion of negative symptoms. And how would you put it together with your paper on reward prediction in the whole negative symptom landscape? That's our challenge when we write a news story that also discusses the study by Green and colleagues [see SRF related news story].
Gold: Well, it’s not that easy, is it? I do have one thought about it, and this may be idiosyncratic to me. My own view is that for a long time people had almost nothing to say about negative symptoms other than that they were really bad. But in terms of what was actually happening, there just wasn’t a whole lot that I read that made it seem like people were trying to grapple with the question, How does this really come about? I think our computational approach was a little bit unusual at the time, where we had a very specific hypothesis about how you could build a brain that would learn from punishing outcomes and not learn as well from positive outcomes, and what that would end up looking like. In a way, that model is highly abstract. The Strauss article is narrower, but it’s motivated in a similar way of, How should one understand this phenomenon that’s been described over and over again in the literature of patients having intact, in-the-moment pleasure but bad prediction of future pleasure and reduced reporting of past pleasure? How could you make that come about, and does it mean that patients really have a pleasure deficit, or do they have a different kind of problem? I think the Strauss argument about beliefs about one’s own experiences fits completely with the Aaron Beck work. His way of framing those issues is entirely consistent with the cognitive model of negative symptoms that Beck proposes and that Michael Green managed to get very prominently into his pathway. These three papers address the role of perceptual, working memory and episodic memory representations, and how those alter motivation or the experience of emotion, and perhaps undermine volition. So, on some level, we’re all barking up similar trees.
Michael’s goal was clearly a different one. It was to try to look at how these things all line up in a mathematical way. Greg’s paper was much more conceptual, and our earlier paper was more data driven and model based. I don’t know if this counts as a renaissance of interest in negative symptoms. But there has been a lot of movement in the field: NIMH had a workshop on the assessment of negative symptoms several years ago. It stimulated a couple of new rating scales that are now being published [see SRF related news story], and I think there’s some pharma interest, though I don’t know for how long.
SRF: It seems to me that the focus on the various cognitive impairments that might give rise to negative symptoms kind of melds two categories of symptoms that characterize schizophrenia. Are they in the same category now?
Gold: I think one has to be careful about how one uses the term "cognitive." There’s a large literature on the straightforward correlation between negative symptom severity and performance on neuropsychological tests, so that would be how most people think about dimensions of the illness. Most of the time, when people are talking about cognition, they’re talking about impaired cognition on conventional measures of ability. Now I’m speaking very loosely, but in general the correlation between negative symptoms and neuropsychological performance is more substantial than that of positive symptoms with neuropsychological performance. But that’s because the positive symptoms correlations are typically very close to zero, whereas the negative symptoms correlations seem to be on the order of magnitude of 0.2, 0.3—sometimes higher. But it’s not the case that they are densely overlapping, because if they were, the correlations would be much higher—these are different things.
Now I think with our two recent papers, or Aaron Beck with his defeatist performance beliefs, we’re not just talking about raw abilities, at least with the reinforcement learning approach. We’re suggesting that there’s a particular process that might be implicated. Now whether that process is related to language ability or visual-spatial ability, any of those, I don’t really know. We have not looked at that issue closely yet. Whether that relates to other aspects of cognition, I just don’t know.
With the defeatist performance beliefs, there’s a paper from the Beck group showing pretty substantial correlations, surprisingly, between their defeatist performance beliefs and how people perform on neuropsychological tests [CIT]. You wonder if they’re assessing the same things as other people have talked about as negative symptoms. For example, does the defeatist performance approach assessment get at people who don’t think they’re going to succeed on these tests and therefore don’t try very hard, because they're frustrating? As opposed to a more straightforward relationship where you might imagine people with worse negative symptoms having impaired frontal cortical function and that’s why they do badly on cognitive tests? I think these terms seem similar, but they’re not, actually.
SRF: Okay, that’s the end of our questions. You’ve been very helpful, and we
appreciate the time you’ve taken.
Gold: Oh, sure. This was fun.