CR Plus, CR adjacent, CR independent - what the heck?

The goal of interventional biogerontology is to find treatments that can extend human lifespan and healthspan. This would allow everyone more quality time with their beloved parents and grandparents, it would reduce social strife (since long-lived people value life more) and would boost the economy because the dependency ratio would decrease (cf. Olshansky et al. 2006). The most promising intervention used to be caloric restriction (CR), i.e. eating less. However, now we have treatments that look kind of different from CR yet are still very similar. So how do we know when we found something truly "new"? Why do we want to know this? And how can we combine treatments and conceptualize the idea of having different - sometimes parallel -  lifespan extending pathways?

So let's do a brief #coffehousenotes post on this topic and please excuse me if this is 50% interesting (hopefully) and 50% a sloppy wild-goose chase.

Abbreviations & glossary

CR, caloric restriction
LS, lifespan
GHRKO, growth-hormone recptor knock-out dwarf mice
FGF21, a fasting hormone most sensitive to protein; FGF21 tg mice overexpress this hormone
Rapa or rapamycin, a life extending drug (at least in mice)
GH/IGF-1, growth hormone/insulin-like growth factor-1 a hormonal axis that is important to lifespan
Myc, a transcription factor involved in cancer and cell growth

What do we want to know?
What story are the mouse models starting to tell us?
Graded CR: what can we learn?
Chronic vs acute CR; cells vs organisms
Addition of two lifespan extending treatments: not so fast, not so easy

What do we want to know?
One of the key questions remaining unanswered, is how closely related are the known mouse models of lifespan extension? Are they all mechanistically overlapping? This seems likely true for most of them. Are they impinging on the same central two or three pathways? This, on the other hand, is less clear. Especially, because it is very hard to tell in the cell signalling business what even is a pathway, a module, what is downstream and what is upstream. Each cell and tissue engages key pathways in different ways across time and space. To some extent the question is so reductionistic that it seems almost naive.

Maybe we can rephrase these questions a bit to make it clearer. The canonical model of lifespan extension is nutrient limitation and in the case of mice this is basically caloric restriction (CR). Presumably, we have mouse models or drug treatments that are closer to CR in their mechanisms and those totally unrelated, as well as everything in-between the two extremes.

As we try to research these treatments the risk-reward changes in various ways. It is still worthwhile to go down the well-trodden path of CR research. We still don't have anything resembling a safe and effective CR-mimetic. However, if we could find a pathway that extends lifespan and is completely independent of CR this would be a revolution because it would give us so many more options to develop future therapies for humans and it is much more likely to provide synergistic or additive benefits with CR. Unfortunately, right now, is my guess, most of the treatments we know I would describe as "CR-adjacent" or "anti-anabolic". When combined they are likely to provide diminishing returns.

It is not even entirely clear if our current research paradigm in mice (KO out stuff, feed the mouse some stuff, etc) is very good at uncovering things beyond basic anti-anabolism and distant CR-mimetics. The ever controversial, ever inspiring Aubrey de Grey is right, at least, when he talks about a different conceptual framework to treat aging.

Figure 1. Different concepts of lifespan extension
A hierarchy of lifespan extending genetic, nutritional and drug treatments. CR per se (including so far unknown genuine CR mimetics), CR-adjacent (we can also call these CR+; perhaps e.g. rapamycin), anti-anabolic and truly novel interventions (which we could call CR-independent). Please note, in reality we may have a gradient with more or less overlap. Findings may be tissue-specific and depend on what you measure, e.g. if you compare inducers, activated pathways, gene expression or outcomes. Treatments that reduce an animal's body size or inhibit pathways involved in nutrient sensing and growth are considered "anti-anabolic".

What story are the mouse models starting to tell us?
Let us look at a recent paper by Gladyshev (Tyshkovskiy et al. 2019) for an incomplete review of lifespan extending paradigms in the mouse. I will give a brief run-down first, with more detailed discussion afterwards:

  • 17-alpha-estradiol - very weak LS extension, category unclear, some evidence that 'feminizing' effects on gene expression are CR-adjacent, and/or dwarf-mode-adjacent, with weak if any benefit for LS (see the Gladyshev paper)
  • Acarbose - very weak LS extension, clearly anti-anabolic, very possibly CR-adjacent
  • Ames dwarf - strong LS extension, clearly anti-anabolic, very, very possibly CR-adjacent
  • CR - caloric restriction is the sine qua non of modern biogerontology
  • DGAT1 deletion - very moderate LS extension, not a widely known intervention, weakly CR-adjacent (e.g. a bit lower IGF-1)
  • EOD (every other day feeding) - modest LS extension, probably anti-anabolic, very, very possibly CR-adjacent
  • FGF21 expression - strong LS extension (not well replicated), clearly anti-anabolic, relationship to CR somewhat unclear
  • GHRKO dwarfism - like Ames dwarfism, with a tendency to be *more* robust. The queen of all interventions holding the world-record for longest lived lab mouse in the world.
  • Little mice - very similar to dwarfism, GH mechanism
  • MR - methionine restriction shows robust LS extension and is clearly anti-anabolic, very possibly CR-adjacent
  • Myc deficiency - strong LS extension (not well replicated), clearly anti-anabolic, relationship to CR somewhat unclear
  • Protandim - very weak LS extension, NOT clearly anti-anabolic, engages a very possibly CR-adjacent mechanism via Nrf2 signalling
  • Rapamycin - moderate to strong LS extension, this is the king of drugs, clearly anti-anabolic, very possibly related to CR or GHRKO (which is quite CR-adjancent itself)
  • S6K1 deletion - moderate LS extension, part of the mTOR sigalling pathway targeted by rapamycin
  • Snell dwarf - analogous mutation to Ames dwarf with tiny differences

Let us pick three examples to discuss in a bit more detail:

FGF21-transgenic mice live much longer than litter mates. While we cannot make the case that FGF-21 is very far from CR, it is probably not quite as CR-adjacent as it seems. FGF21-Tg mice are smaller than normal mice, have reduced IGF-1 and insulin levels and, e.g., inconsistently higher adiponectin, which is strong evidence for a CR mimetic effect (Zhang et al. 2012). Just like dwarfs and CR mice these FGF21-Tg mice also engage the recently uncovered H2S pathway and FGF-21 seems to be positively regulated by PGC1a & ATF4 which are often upregulated in CR-adjacent models. Despite these similarities there is quite a bit of evidence that CR, GHRKO and rapamycin do NOT induce FGF-21. I find this entirely unbelievable in a good way (awe of the "wow"-variety). Again just to reiterate. It is a picture-perfect CR mimetic hormone that is not induced by CR or most CR-adjacent treatments. As it turns out, FGF-21 seems to be a response to protein starvation and carbohydrate abundance. Sadly, strong protein restriction per se is toxic for some reasons (except methionine, see below), or only modestly beneficial when implemented at lower intensity, and the FGF-21 data in humans has red flags written all over it (it's up in mitochondrial disease and during obesity). The optimist in me thinks that FGF-21 can be used to boost a moderate CR state or another drug's efficacy. So it will be very interesting to see how this one shakes out. Still, it is obviously an intervention of the nutrient sensing/anti-anabolic variety and a distant relative of CR (much more so downstream than upstream).

Methionine restriction (MR) is the least toxic of the protein restriction paradigms and in fact quite on the salubrious end of the distribution. It is indeed really healthy for rodents. Interestingly, MR engages  similar pathways to FGF-21 transgenic over-expression. Same high H2S, low IGF-1, low insulin, low body-weight quadrifecta. If it is not clear yet where this is going, and I am not sure if people made this connection before, but it seems to me that MR could be the best way to induce FGF-21 expression. (Perhaps we should emphasize the 'could' because there are some contradictory reports out there, at least for hepatic FGF-21 expression, see e.g. the Gladyhshev paper and compare with the above linked work by Lees et al. 2014) Maybe we can conclude that MR and FGF-21 are in the same pathway with FGF-21 tg being an MR-mimetic. Both then are obviously anti-anabolic and in many ways CR-adjacent.

Myc deficiency (Myc +/- haploinsufficiency; Hoffman et al. 2015) extends lifespan in mice. It is a sensible choice to take a look at Myc becaue it is obviously pro-growth and a known oncogene. Just knowing these facts it seems like a good hunch to go for reduced Myc. Is this close to CR or not? At first glance it does not seem very CR-adjacent because no one was seriously looking at Myc downstream of CR, or at least not many were. I don't know if CR mice have reduced myc activity. But even if they don't, does this mean anything if the mice are smaller and have reduced IGF-1 anyway, thus are a strong pheno-copy of CR? As far as I am aware, no one tested H2S signalling in these mice and I can't seem to find any insulin data in the paper.

Although the authors say: "In contrast to observations in other longevity models, Myc+/− mice do not show improvements in stress management pathways". This statement by itself is not enough to prove that it is not CR-adjacent, since I've seen similar results before in models that I would consider reasonably CR-adjacent. It is certainly one line of evidence that myc deficiency may be a bit farther from CR than e.g. dwarfism. In a similar fashion we could highlight the unchanged adiponectin levels, which leads us to a tentative conclusion that xenobiotic metabolism and adipokines may be the most variable across models.

Also in many other ways, few of these mouse models seems to break the mold. For example, CR, GHRKO and Myc +/- all have reduced mTOR signalling in the liver. A bit surprisingly mTOR signalling was reported as unchanged in FGF-21 tg. So far I've never felt the urge to double check if MR decreases mTOR signalling, which I assumed (or even incorrectly recalled) to be true without a reference. I just checked it because I don't want to say anything wrong. At a quick glance it seems that it does reduce mTOR signalling (Lees et al. 2016).

Another approach is of course to compare the transcriptomic response of these mouse models as the Gladyshev group did (Tyshkovskiy et al. 2019). Here it comes as a bit of a surprise that Myc and rapamycin are less similar to CR and dwarfism than is metformin (which has only tiny effects on lifespan!) On the other hand, CR and dwarf models are similar which fits our intuition and the hormone data. This teaches us that we shouldn't take transcriptomics too seriously, although, we can't ignore it. Such a cautious stance makes biologic sense since there is a lot of regulation that takes place post-transcriptionally and is not captured by basal hepatic gene expression.

Graded CR: what can we learn?
This paradigm has been pioneered by John Speakman even though the idea is obviously not new. Some simple graded approaches have also been applied in GH/IGF1 research, e.g. GH dwarfism vs normal vs GH over-expression.

Does the graded CR paradigm point the way towards bona-fide aging- and CR-related mechanisms? Sure, it is helpful but not magic. Graded CR can only help us distinguish bona fide mechanisms of graded and chronic CR mechanisms from those that are not. The 'not' group encompasses several things, most importantly, true mechanisms of CR that *are* associated with lifespan extension but not actually engaged in the chronic or graded mode (de facto CR-adjacent). In contrast, the method unfortunately also shows us the "true" mechanisms of CR that are *not* associated with lifespan extension under any circumstances. How come you ask?

Figure 2. Graded CR does not imply (although it suggests) graded underlying mechanisms.
Just because the lifespan response is graded, does not mean the mechanisms have to be graded. Mechanism here refers to changes in gene expression, phosphorylation state of key signalling pathways (e.g. mTOR) and hormonal and circulating *kines and so on (e.g. IGF-1, chemokines, myokines, etc).
Let us look at three possible examples leading to the very same outcome. 1) 10% CR engages pathway A, whereas only under 20% CR is pathway B engaged and under 40% CR pathway C is engaged. None of these pathways changes in a graded manner. 2) Standard assumption, simplified a bit, CR engages pathway A and this pathway increases in intensity going from 10 over 20 to 40% CR. 3) Is a combination of 1+2 and I would consider this the most plausible scenario. This is still ignoring more complicated cases were some pathways may be fully turned off and replaced by others, even if they contribute to lifespan extension under a certain level of CR!

To expound on the above concepts:

1) True mechanisms of CR that are *not* associated with lifespan extension under any circumstances, could mean mechanisms that are necessary to survive under CR conditions, usually in the wild. So this could be e.g. increased activity and foraging behavior (hunger -> search for food). Similarly, this could include "mechanisms" or outcomes that can be actively harmful at times, e.g. decreased wound healing or leukopenia (although, we can't even confidently rule out the latter as causal!)
2) True mechanisms of CR that *are* associated with lifespan extension but not actually engaged in the chronic or graded mode (de facto CR-adjacent), could mean mechanisms that are somehow 'capped'. Imagine you have two growth limiting pathways. If you engage both of them fully the organism dies or struggles to maintain homeostasis in the wild. CR engages both by 50% or at different levels of restriction. A CR-adjacent treatment could just engage one of them, but instead by 90% of the maximum.

Not quite the same thing as (2), but apropos 'capped', the GH/IGF-1 axis comes to mind. It is likely that dwarfs engage this axis more strongly -- this is extremely plausible when you look at actual circulating IGF-1 levels. These are low in CR, much lower in dwarfs. Admittedly, there is another conceptual difficult: it is hard to draw the line between CR-mimetic and CR-adjacent. But remember, we are looking for a *heuristic* to understand and plan aging research and not THE answer to all our problems!

Chronic vs acute CR; cells vs organisms
Similar to graded CR, we can look at gene expression changes in starved cells, tissues or organisms as compared to "real" CR. This is not the same because starvation is a short term response and CR is generally defined as long-term, "moderate starvation" with optimal nutrition, so to say. Well basically, it is NOT starvation but obviously it is also related.

One could make the mistake of saying that chronic CR is the only admissible gold-standard and one would be right, if we were ONLY looking for strong CR-mimetics and very, very CR-adjacent pathways. As explained above, starvation could activate several life-extending pathways that are simply not compatible with long term survival taken together. However, each one of them independently could be useful to extend lifespan, maybe even with fewer side-effects than CR. This is plausible because metabolism and gene expression is organized in modules. It is likely that ALL processes activated by CR would be -- purely statistically speaking -- enriched in processes that are good for lifespan. However, only experiments can provide the final evidence for these ideas.

Studying acute CR is simply higher reward, higher risk than studying gold-standard chronic CR, but it is also cheaper. For example, I recall activation of several lysosomal transcription factors (generally higher expression) during starvation or fasting with no data from bona fide CR. This is not worthless, this is promising for the biogerontolgist. 

Perhaps even more importantly, one has to think about metabolic and diurnal variability. Maybe acute starvation *is* somewhat representative of what goes on during the period between meals. We know from time restricted feeding and EOD studies that there is something there.

Figure 3. Chronic vs acute CR
Some mechanisms are maintained during chronic and acute CR (top row; e.g. reduced mTOR signalling). Here the screening using acute CR as a model (often in vitro) will save time and money. Some mechanisms are lost during chronic CR, or put another way, up during acute and unchanged (unch.) during chronic CR (middle row; e.g. perhaps ketogenesis for acute starvation and to some extent thyroid and other hormones). If we know which mechanisms behave that way, we can uncover pathways that are beneficial to lifespan and additive to CR (if they were lost due to a fitness disadvantage that we can mitigate) or we can go down a blind alley looking at harmful mechanisms. Finally, some mechanisms are unchanged during acute CR and only turned on during chronic CR. These we can only uncover using expensive long-term studies. It is surprisingly difficult to come up with an example of the latter. I think (although I am not 100% confident) that Nrf2 induction is well documented in long-term CR but - in contrast to autophagy and mTOR signalling - not well documented in the cheapest model of acute CR (serum starved cells).

Addition of two lifespan extending treatments: not so fast, not so easy
If two treatments fail to provide additive benefits on lifespan extension it does not mean they engage the same pathway, although it can be considered evidence for this proposition especially when backed up with molecular biology. One would think that if you combine two mostly independent life-span extending treatments this would lead to additive benefits. Yes, in theory. In practise, it is a bit less trivial.

What if a novel pathology X is limiting lifespan of mice around 1300-1500d, and only at such extreme ages, and neither of your independent LS extending treatments affects it? In this case you would still notice that your LS extending treatments are additive if you were to combine two 'weak' versions of them (imagine, conceptually, giving two drugs at half the dose). We know mice tend to have sometimes rather specific issues leading to death like nephropathy or lymphoma, depending on the strain. A drug could slow aging but fail to reduce cancer rates, and the lifespan of mice would be capped at a certain age, simply due to those pesky tumors. Yet, in humans such a drug or drug combination might still save billions.

What if both of your treatments have moderate toxicity at a target organ, especially the liver? This may not limit lifespan under normal circumstances, but what if you combine drugs, what if CR decreased a certain metabolic activity in the liver (maybe not even directly, but due to changes in tissue size and plasma levels etc)? We know mice have rather sensitive livers.

Let us look at one specific example. The study by Yimin et al. (2018) from the Bartke group is interesting as it pertains to the combination of dwarfism + rapamycin. If we were to trust the Gladyshev transcriptomic data, we might start with the working hypothesis that they are independent and should be additive. If like me you consider the anti-anabolic phenotype informative you'd speculate that they are additive but may provide diminishing returns. In reality, it's even worse:

"We found that rapamycin extended life span in control normal (N) mice, whereas it had the opposite effect in GHR-KO mice. In the rapamycin-treated GHR-KO mice, mTORC2 signaling was reduced without further inhibition of mTORC1 in the liver, muscle, and s.c. fat. Glucose and lipid homeostasis were impaired, and old GHR-KO mice treated with rapamycin lost functional immune cells and had increased inflammation. "

These results provide evidence for certain assertions, but ultimately it is hard to discount the alternative hypothesis (no this is not goalpost shifting). If we stipulate that there are three effects of mTOR inhibition in a Gedankenexperiment. Increased autophagy, immunosupprsesion + muscle catabolism, decreased translation. We could trivially construct a case that shows an outcome of "no additive benefit" due to immunosuppression even if two of three mechanisms are additive and beneficial (both conceivably counteract protein aggregation, for example, and could reduce the incidence of mtDNA deletion clonal expansion).

I just think we need to dig deeper. Let us just highlight some issues in no particular order. The authors used an "injection protocol of rapamycin". Even if they have a vehicle control, and it is not unusual, I already dislike this protocol. No one is ever going to inject rapa in humans and the kinetics are different from oral microencapsulated rapamycin. Could make a difference if very low trough mTOR signalling is an issue in these mice.

"End-of-life histology showed that GHR-KO mice with prolonged rapamycin treatment had an accumulation of pigmented alveolar macrophage-like cells in the lungs and pigmented macrophage-like cells in the spleen"

So, IF this were a life-limiting pathology in the combined model, and the only one, with some trivial solution. How does it affect our conclusion? "Rapa+GHRKO magically becomes additive if you give a drug limiting this immune response"? Additive or not additive is not a black and white answer. We can say it is NOT additive under very specific circumstances, maybe it is additive under a different protocol? Again note the astonishing decrease in lifespan in the combined group. This is redolent of toxicity and massive toxicity at that. Why do we expect life-shortening a priori if we do not assume some limiting toxicity ? It would be plausible to expect a null effect on lifespan.

I am not aiming for apologism per se. There is no doubt at all that this study is very well done and very, very bad news for the optimists (who want more additive treatments of course). No need to be shy about the truth. 

"percentage of fat mass (Fig. 3B) were lower than in vehicle-injected...Prolonged rapamycin treatment elevated fasting glucose levels..the mice treated with rapamycin became insulin resistant"

We know that in GHRKO the fat depots are reprogrammed to secrete beneficial hormones. A reduction could be bad. Some strain-specific issues with rapamycin and glucose handling are also well-reported. One could speculate that this has an effect. Again, from a translational perspective it is not clear this is even relevant because we would much rather combine moderate reduction in the GH/IGF-1 pathway with moderate doses of rapamaycin. Extreme GH deficiency has a number of side-effects in humans. 

So the authors obviously speculate along the same lines here: "GHR-KO mice lost much of their functionally unique and “healthy” fat depots, which was accompanied by detrimental alterations in their circulating lipid profile".

Finally, the model the authors favor is this one. As per Fig. 2 rapamycin did not affect mTORC1 signalling in liver, muscle and subcutaneous fat. However, it did reduce insulin stimulated mTORC2 signalling (pS473-AKT). If we posit that mTORC2  inhibition is an undesirable effect to begin with, this could explain the toxicity we see. Theoretically if we had an mTORC1 specific inhibitor we might see further benefit in GHRKO. This is an interesting working hypothesis. I can see Dudley Lamming nodding quietly "I told you so".

As we can see from the above discussion, mouse studies can help us to assign probabilities to different hypotheses. We can try to answer, for example, if different treatments are just CR+ and hence won't be additive or if they will. However, we have to consider that mouse studies are expensive and thus we cannot use them to probe sufficient 'space' that contains a valid solution to explain and reconcile all distinct data points. Even with the above studies we are still left asking: why does rapamycin show such a robust dose-response in the NIA ITP but it fails in GHRKO mice? Why do CR and dwarfism look so similar when you add them to the mix, as if they were operating in the same pathway, yet there are no successful adult-onset GH deficiency studies and plenty of positive CR and rapamycin studies started late? Dose, strain, toxicity or real lack of synergy?

By the way, I like to phrase it this way, because it reminds of another fundamental issue in aging research: the experimental bottleneck. Until we decide to throw real money at high-throughput and medium through-put screens we will never get a solid answer and figure out what works, or what works together and under which circumstances. The experimental bottleneck means we have sparse data that often cannot be overcome by theory, theorists or machine learning -- despite what is often hoped.

Now on to some predictions, so I can embarrass myself in ten years when I read this post. I am somewhat confident that rapamycin + dwarfism and CR + dwarfism* are additive when we use the right protocol, although, with diminishing returns. I am much more confident that rapamycin will be additive to moderate CR or moderate IGF-1 deficiency.

*I am aware of the literature and I interpret it as somewhat supportive of this assertion

Do I think we know treatments that are genuinely distinct from CR? Not sure about that. I would put most of the known strategies at "very related but distinct". Harking back on the introduction, maybe all the known treatments do fit two or three major pathways, but then still I'd rather think of interconnected partially overlapping modules than pathways per se.

Although, CR clearly reduces the senescent cell burden, I would say that senolytics are quite different mechanistically. It seems this is the first therapy having a small but consistent lifespan benefit that does NOT engage CR-adjacent pathways (IGF-1, body size, mTOR, etc).

Will we discover a life extending treatment that is not anti-anabolic any time soon? Will we discover any useful gain of function mutations that are not anti-anabolic? Is there any hope for drugs that do not fit this mold in mice? Or maybe we already discovered one of these and it is not well validated? Maybe not, or not in a simple way, if cancer is a bottleneck - but then boosting DNA repair would be a hope. Oh, god, this, however, seems very difficult! Since these systems are beyond intricate, although perhaps a bit more imperfect in the mouse than in humans. So, yeah, let me end by saying that finding a life extending treatment that is not anti-anabolic would greatly advance the field, but this may await more funding or improved protocols. Which protocols? At least superficially I liked the idea of 'unmasking', which I think I am stealing from Aubrey or his collaborators; slow down cancer sufficiently to see if you can get something on top of that. It would be also nice if we could start with aged, already super-long lived mice and then perform a medium throughput screen, either small groups for late-in-life functional rejuvenation or LS extension, or just screen for improved surrogate endpoints; which I don't trust that much.

References (selection)

Fang, Yimin, et al. "Effects of rapamycin on growth hormone receptor knockout mice." Proceedings of the National Academy of Sciences 115.7 (2018): E1495-E1503.

Olshansky, S. Jay, et al. "In pursuit of the longevity dividend: what should we be doing to prepare for the unprecedented aging of humanity?." The Scientist 20.3 (2006): 28-37.

Tyshkovskiy, Alexander, et al. "Identification and application of gene expression signatures associated with lifespan extension." Cell metabolism 30.3 (2019): 573-593.

Zhang, Yuan, et al. "The starvation hormone, fibroblast growth factor-21, extends lifespan in mice." elife 1 (2012): e00065.

Hofmann, Jeffrey W., et al. "Reduced expression of MYC increases longevity and enhances healthspan." Cell 160.3 (2015): 477-488.