I needed to disable self-signups because I’ve been getting too many spam-type sign-ups lately. Please contact me directly if you want membership on this forum. Thanks.

Forum Navigation
Please to create posts and topics.

Expression of Vitamin A-related genes increases with BMI

I think humans would be no different than any other animal: Most toxins get stored organs and bones. Thats where you would find most heavy metals in animals so I assume it would be the same for humans.

[There is a summary at the end of the post]

After inspecting the data some more I discovered that some of the samples contained negative values, which I assume means that the samples did not contain quantifiable mRNA. Including these negative values in the initial analysis therefore led to falsification of some results. The only results affected by this issue are those for CYP26A1 (detectable in 76% of control samples), CYP26B1 (84%), CYP26C1 (24%) and CYP7A1 (92%), the remaining results should be correct.

Especially frustrating is the fact that CYP26C1 was detectable in so few samples. Some possible explanations are

  • a) Since CYP26C1 shows 10x greater affinity for metabolizing 4-oxo-RA than CYP26A1, less enzyme may be required to clear atRA, and therefore even though the amount of CYP26C1 is too small to be quantified using microarrays, it could still be the main enzyme responsible for clearing atRA in vivo
  • b) One of the other CYP enzymes is the primary enzyme for clearing atRA
  • c) Retinoic acid must leave the liver and get metabolized somewhere else before it can be eliminated

To determine whether option c) is plausible, I analyzed the expression of various genes in adipose tissue taken from 770 Finnish males (Civelek, Wu et al. 2017). Surprisingly, CYP26C1 was quantifiable in 76% of control samples whereas CYP26A1 was only quantifiable in 25% of samples, which is almost the exact opposite of the distribution in liver. Additionally, unexpectedly, LRAT was not quantifiable in any of the samples.

I then found a study comparing the pharmacokinetics of 13-cis-RA (Accutane) in patients with different CYP variants (Veal, Errington et al. 2013). The study showed that patients with the CYP3A7*1C (present in 2.8% of the group) and CYP2C8*4 (present in 6.2%) polymorphisms had 2x and 0.6x more 4-oxo-13-cis-RA in plasma, respectively, compared to patients with the more common variants. CYP3A7*1C has been associated with a 54% reduction in urinary estrone glucuronide and significantly increased mortality in various cancers including breast and lung cancer (Johnson, De Ieso et al. 2016).

Since CYP3A7 catalyzes the hydroxylation of estrone, and urinary estrone glucuronide was decreased in subjects with CYP3A7*1C, it seems likely that this variant is more efficient at hydroxylating estrone and therefore will probably also be more efficient at hydroxylating RA. Since CYP3A7 preferentially forms the (4S)-OH-RA enantiomer, and plasma 4-oxo-13-cis-RA was increased in subjects with the more efficient CYP3A7*1C variant, I think we can conclude that an increased production of (4S)-OH-RA leads to increased formation of 4-oxo-RA.

Conversely, CYP2C8*4 has been associated with decreased enzyme function (Aquilante, Niemi et al. 2013). Since CYP2C8 also preferentially forms (4S)-OH-RA, CYP2C8*4 therefore forms less (4S)-OH-RA, and since less 4-oxo-13-cis-RA was produced in subjects with CYP2C8*4, this confirms that decreased production of (4S)-OH-RA leads to decreased formation of 4-oxo-RA.

A recent study by the same team that previously studied CYP26C1 noted that CYP3A7 constitutes 30–85% of all CYPs in human fetal livers, and suggested that CYP3A7 (not CYP26A1) is the enzyme primarily responsible for clearing atRA in human fetal livers (Topletz, Zhong et al. 2019). I disagree with this interpretation, and instead believe that these findings add more support to the theory that (4S)-OH-RA is recycled whereas (4R)-OH-RA is eliminated. Since embryos apparently require retinoic acid to grow, it is therefore expected that primarily (4S)-OH-RA is formed, specifically because it cannot be eliminated.

For these reasons, it seems like CYP3A5 is now the only remaining candidate for RA-clearing enzymes in the liver, since it is also the only remaining enzyme that has been demonstrated to preferentially form (4R)-OH-RA (except CYP26C1). However, CYP2C9 is the second-most expressed CYP enzyme in human liver, shows activity towards retinoic acid and is known to be stereoselective, which is why I included it in the first analysis with a question mark. The SSRI R-fluoxetine was reported to be a 5x greater inhibitor of CYP2C9 than S-fluoxetine (Schmider, Greenblatt et al. 1997), and R-lansoprazole (but not S-lansoprazole) activated both CYP2C9 in human liver microsomes and recombinant CYP2C9 (Liu, Kim et al. 2005). Since CYP3A5 was progressively more downregulated in adipose tissue, but not in liver as BMI increased, while CYP2C9 expression increased in both liver and adipose tissue, CYP3A5 may be the main hepatic RA-clearing enzyme whereas CYP2C9 could be the main RA-clearing enzyme in adipose tissue. Formation of (4R)-OH-RA by CYP2C9 remains to be demonstrated.


CYP26C1 was only detectable in 25% of samples in the initial analysis, raising doubts about whether it is in fact the primary enzyme responsible for clearing retinoic acid. I found more evidence to support the 4R/4S theory, however it’s starting to look like CYP3A5, and not CYP26C1, may be the primary enzyme responsible for clearing retinoic acid in the liver. Nonetheless CYP26C1, possibly together with CYP2C9, may still be responsible for clearing retinoic acid in adipose tissue. Additionally, CYP3A7 may play a role in embryogenesis.

Quote from tim on October 8, 2021, 5:53 am

If anyone has any evidence that significant amounts of vitamin A are stored in human adipose tissue please post it. Both tallow and lard are low in vitamin A despite cattle and pigs consuming diets that are not low in vitamin A.

Obesity and fatty liver disease is associated with decreased liver retinol stores and higher serum retinol levels in mice. Changes in the expression of important retinol metabolizing enzymes are possibly due to the effects of metabolic dysfunction on vitamin A metabolism. Weight loss and lack of appetite are two of the most important symptoms of Hypervitaminosis A.


I wouldn’t say lard is low in vitamin A, one study found that it contained about 550µg/kg (Ayuso, Óvilo et al. 2015). Of course, it’s also widely accepted that 50–80% of total retinol is stored in HSCs in humans. However, we need to be very careful distinguishing between the effects of retinol and retinoic acid. Retinol directly induces lipogenesis via STRA6/STAT5/PPARG whereas retinoic acid induces beta-oxidation via PPARD and possibly PPARA, both of which form heterodimers with RXR.

It has been shown that lipophilic toxins like PCBs are exclusively stored in lipid droplets, and some of these lipid droplets end up in adipocytes (Bourez, Le Lay et al. 2012). Additionally, animals consistently lose roughly 25% of their body weight and even more body fat after removing Vitamin A from the diet. We also know that Vitamin A supplementation for a few months does not cause weight gain or loss in humans. For all these reasons, my working theory is that lipid droplets containing Vitamin A are first distributed to HSCs, and only when the liver can’t accommodate any more retinol the excess is released into the bloodstream and taken up in adipose tissue elsewhere. Plasma RBP4 was also found to correlate with body weight (Wang, Huang et al. 2020). 


Aquilante, C. L., M. Niemi, L. Gong, R. B. Altman and T. E. Klein (2013). "PharmGKB summary: very important pharmacogene information for cytochrome P450, family 2, subfamily C, polypeptide 8." Pharmacogenetics and genomics 23(12): 721-728.

Ayuso, M., C. Óvilo, A. Fernández, Y. Nuñez, B. Isabel, A. Daza, C. J. López-Bote and A. I. Rey (2015). "Effects of dietary vitamin A supplementation or restriction and its timing on retinol and α-tocopherol accumulation and gene expression in heavy pigs." Animal Feed Science and Technology 202: 62-74.

Bourez, S., S. Le Lay, C. Van den Daelen, C. Louis, Y. Larondelle, J.-P. Thomé, Y.-J. Schneider, I. Dugail and C. Debier (2012). "Accumulation of polychlorinated biphenyls in adipocytes: selective targeting to lipid droplets and role of caveolin-1." PloS one 7(2): e31834-e31834.

Civelek, M., Y. Wu, C. Pan, C. K. Raulerson, A. Ko, A. He, C. Tilford, N. K. Saleem, A. Stančáková, L. J. Scott, C. Fuchsberger, H. M. Stringham, A. U. Jackson, N. Narisu, P. S. Chines, K. S. Small, J. Kuusisto, B. W. Parks, P. Pajukanta, T. Kirchgessner, F. S. Collins, P. S. Gargalovic, M. Boehnke, M. Laakso, K. L. Mohlke and A. J. Lusis (2017). "Genetic Regulation of Adipose Gene Expression and Cardio-Metabolic Traits." Am J Hum Genet 100(3): 428-443.

Johnson, N., P. De Ieso, G. Migliorini, N. Orr, P. Broderick, D. Catovsky, A. Matakidou, T. Eisen, C. Goldsmith, F. Dudbridge, J. Peto, I. Dos-Santos-Silva, A. Ashworth, G. Ross, R. S. Houlston and O. Fletcher (2016). "Cytochrome P450 Allele CYP3A7*1C Associates with Adverse Outcomes in Chronic Lymphocytic Leukemia, Breast, and Lung Cancer." Cancer research 76(6): 1485-1493.

Liu, K. H., M. J. Kim, W. M. Jung, W. Kang, I. J. Cha and J. G. Shin (2005). "Lansoprazole enantiomer activates human liver microsomal CYP2C9 catalytic activity in a stereospecific and substrate-specific manner." Drug Metab Dispos 33(2): 209-213.

Schmider, J., D. J. Greenblatt, L. L. von Moltke, D. Karsov and R. I. Shader (1997). "Inhibition of CYP2C9 by selective serotonin reuptake inhibitors in vitro: studies of phenytoin p-hydroxylation." Br J Clin Pharmacol 44(5): 495-498.

Topletz, A. R., G. Zhong and N. Isoherranen (2019). "Scaling in vitro activity of CYP3A7 suggests human fetal livers do not clear retinoic acid entering from maternal circulation." Scientific Reports 9(1): 4620.

Veal, G. J., J. Errington, S. E. Rowbotham, N. A. Illingworth, G. Malik, M. Cole, A. K. Daly, A. D. J. Pearson and A. V. Boddy (2013). "Adaptive Dosing Approaches to the Individualization of 13-<em>Cis</em>-Retinoic Acid (Isotretinoin) Treatment for Children with High-Risk Neuroblastoma." Clinical Cancer Research 19(2): 469-479.

Wang, X., Y. Huang, J. Gao, H. Sun, M. Jayachandran and S. Qu (2020). "Changes of serum retinol-binding protein 4 associated with improved insulin resistance after laparoscopic sleeve gastrectomy in Chinese obese patients." Diabetology & Metabolic Syndrome 12(1): 7.

ggenereux, Jenny and 3 other users have reacted to this post.

@johannes2 thanks for your interesting posts and your chapter too. 

It’s all very interesting. The correlation of vA supplementation and sudden weight gain was for me so strong that I cannot believe anything other than (for me) excess vA equals weight gain. This was combined with a loss of appetite! Very weird but true. Even my husband who was a great ‘eat less, exercise more’ believer realised something very peculiar was going on. 

I’m glad to say after 3 years I’ve lost 18 lbs and feel much more like me again. 12 more to go. 

kathy55wood has reacted to this post.

Lard and tallow are so low in vitamin A that most food databases don't even bother to measure it. Assuming 550 mcg/kg is close to average then it is 100x lower in vitamin A than pork liver and more than 10x lower in vitamin A than butter. From a dietary perspective if one uses 40 grams to cook with that amounts to 22 mcg which is insignificant.

I use lard every day and my progress depleting vitamin A does not appear to have been hampered, my health has consistently improved since I began a low vitamin A diet 2.5 years ago.

Vitamin A in the liver is stored in tiny amounts of lipid. I've wondered if the depletion of liver stores of vitamin A also releases other fat soluble toxins potentially explaining some of the symptoms experienced on a low vitamin A diet.

Additionally, animals consistently lose roughly 25% of their body weight and even more body fat after removing Vitamin A from the diet.

Weight loss in healthy animals with a previously normal weight?

zerocool has reacted to this post.

@johannes2 I’ll be very interested if you look at other genes at some point. Early on I looked at beta carotene genes in some detail and it seemed to me that as people got vA toxic they lost the negative feedback loop that protected them from absorbing too much beta carotene. This involved SCARB1, BCO1 and ISX. Then if they were poor converters (slow BCO1)  the beta carotene would accumulate. People seemed to become ‘super’ absorbers at some point and loose the loop that controlled this.

This is why people who always say you can’t overdose on betacarotene as the body has a protective mechanism aren’t necessarily correct. It does have a mechanism, but what happens if that goes wrong? I suspected that vA toxicity made this go wrong in some way. It was only a theory based largely on anecdotal evidence. The experiments I wanted to see hadn’t been done (or I couldn’t find them) and my attention went onto other aspects of the detox. 


‘Thus, efficiency of carotenoids absorption is controlled by a negative feedback, in which ISX suppresses the intestinal expression of SCARB1 and the vitamin A forming enzyme BCO1 via retinoic acid signaling directly linked with the vitamin A demand of the organism’ 



‘We identified the ISX protein as a critical molecular mediator of this cross talk between diet and immunity. This transcription factor regulates vitamin A production from dietary precursor molecules. Loss of this control in mice disrupts vitamin A homeostasis and impairs immunity.’



Ourania and kathy55wood have reacted to this post.

So I have been working on calculating the significance of individual values, which is a real challenge because I don’t have any clue about statistics. If I’m doing something wrong, please let me know. I did a t test for every BMI range compared to control, and marked the values that were significant at p < 0.05 (*), p < 0.01 (**) and p < 0.001 (***). Additionally, I did a second t test for every BMI range compared to the remaining samples, excluding controls, and marked the values that were significant at p < 0.05 (†), p < 0.01 (††) and p < 0.001 (†††). These two statistical tests are completely independent from the other values displayed in the charts, and were performed using the raw data and not the transformed data for the charts. My code calculates the t values and then I used a website to calculate the corresponding p values, with N1 + N2 - 2 degrees of freedom. Additionally, I removed the smoothing that was present due to the way I had categorized the samples, and all values are now in exactly one group, i.e. the BMI > 25 group becomes BMI 25–30, etc.

While not directly related to Vitamin A, I think this chart is a very accurate representation of the changes the body undergoes while gaining weight, and it seems to match up very well with the observed progression of weight gain. At BMI 20–25 the body mainly burns fat, and this process is apparently mediated primarily by PPARD and not PPARA, while PPARG is normal. Expression of PPARA and PPARD differed significantly compared to the higher BMIs. Expression of PPARG, which induces lipogenesis, appears to consistently increase until BMI 50, and it was significantly increased compared to control at BMI >60. At BMI 25–30 the body first begins limiting lipolysis by downregulating PPARD, possibly due to increased hypoxia. To resolve the hypoxia, PPARA is upregulated from BMI 35–45, possibly to induce beta-oxidation in different kinds of tissue. This upregulation may cause some people to transition to NAFLD and NASH. When that also fails to resolve the hypoxia, eventually the body appears to limit any kind of beta-oxidation at around BMI 50. PPARA and PPARG were significantly downregulated in the BMI >50 groups.

To provide another example, even though CYP7A1 expression was increased by up to 150%, and expression of SQLE by up to 100%, these changes were apparently not statistically significant. Unexpectedly CYP7B1, which was only downregulated up to 20% compared to control was apparently statistically significant.

Quote from Jenny on October 9, 2021, 2:08 am

@johannes2 I’ll be very interested if you look at other genes at some point. Early on I looked at beta carotene genes in some detail and it seemed to me that as people got vA toxic they lost the negative feedback loop that protected them from absorbing too much beta carotene. This involved SCARB1, BCO1 and ISX. Then if they were poor converters (slow BCO1)  the beta carotene would accumulate. People seemed to become ‘super’ absorbers at some point and loose the loop that controlled this.

I calculated the expression of the genes you mentioned, and to be honest I can’t really say what the results mean. While at least BCO1 and ISX are apparently involved with weight gain somehow (?) I’m not sure I see a correlation between ISX and BCO1/2. I’ll let you draw your own conclusions, I wanted to just post the data so you could look at it without me speculating too much. I also didn't yet have time to read the studies you posted, which I want to do before I start interpreting the data. ISX was only quantifiable in 50% of control samples and the distribution appears to be associated with BMI, and possibly AHRR.

Since some carotenoids are AHR receptor ligands, I also looked at AHR-related genes, namely ARNT (which forms a heterodimer with AHR) and AHRR, which is a corepressor for AHR (by forming a heterodimer with ARNT that does not induce transcription). I didn’t immediately see any correlation with BCO1 here either.

I’ll make a post with some more charts in the next few days, hopefully with more detailed annotations, I just wanted to briefly introduce the new significance tests and ISX-related results here.

Edit: Keep in mind that BCO1 may be more active in small intestine and skin than in liver, and that these results are from liver samples. I'll post results from adipose tissue later.

Ourania has reacted to this post.

@johannes2 thank you for data. It would look to me that ISX expression reduces at high BMI? ISX controls the absorption and conversion of betacarotene from food. If this is not expressed normally, then the negative feedback loop that should stop us absorbing too much betacarotene when we have plenty of vA, would not work correctly. This would correlate with what we see I think if I’m interpreting it correctly. As the liver becomes sicker (as would happen at higher BMI) we loose control of our beta carotene absorption and this adds to our problems. High vA vegetables become toxic. 

‘ISX suppresses gene expression of the scavenger receptor class B type 1 (Scarb1) and the β-carotene-15,15′-dioxygenase (Bco1), which encode proteins that respectively mediate the uptake of carotenoids and their conversion into retinoids (810). This negative feedback regulation controls the utilization of dietary BC for retinoid production in mice depending on demand and availability (9, 10).’ 

(from paper linked in above comment) 

If ISX expression is compromised it can be seen how things go awry. 

Ourania and kathy55wood have reacted to this post.

So in other words: The higher the BMI of an individual the greater the harm caused by high beta-carotene foods?

kathy55wood has reacted to this post.

I don’t know. It’s all theory. The data that Johannes produced seems to say that less ISX is expressed at higher BMI (unless I’ve interpreted wrongly). ISX suppresses the proteins that absorb and convert betacarotene to retinal. Lack of ISX would suggest that this negative feedback is working less well. Less suppression is taking place which means more absorption and conversion. I would suspect it’s to do with liver health as well as BMI, but that’s just a guess. I think that vA toxicity messes up this helpful negative feedback loop that prevents too much betacarotene being absorbed. So yes I’d say the sicker the person (BMI and/or liver health) the more harm caused by betacarotene foods. This is all theory and largely from anecdotal observations, but correlates with my experiences. Clearly the idea that people can’t become toxic from carotenoid intake, due to the negative feedback loop, is incorrect. 

Ourania and kathy55wood have reacted to this post.

Introducing End S***ty Gene Software

Because almost all apps for analyzing gene expression data only run on Windows, and because manually generating the charts was becoming tedious, I updated my program which calculates the results to also generate the charts, and to additionally calculate the significance of the results. Even though I hope to publish a study about this at some point, I’ve decided to make this app available for use by people on this forum, and I’m calling it “End S***ty Gene Software” (ESGS) for now. You will need a Mac to run ESGS. There are screenshots of ESGS further down in this post.

My goal was to make using ESGS as easy as possible. First, download datasets from the website refine.bio. You can download the ones I compiled, which are confirmed to work with ESGS, here (liver) and here (adipose tissue). Then, extract the ZIP files, open ESGS, click the “Open” button and choose the folder that contains the data. This folder should contain a file called “aggregated_metadata.json”. After opening the folder, there should be a “Ready to analyze package” message visible. Finally, enter a list of approved gene symbols (or Ensembl IDs) and click the “Analyze” button. If the checkbox is checked, the BMI ranges will be smaller, e.g. 25–27.5 instead of 25–30. If the app freezes for more than a few seconds after clicking the “Analyze” button, you should assume that a problem occurred and force quit the app.

Download for MacOS

I tried to incorporate all the available data into a model of weight gain based on BMI. The observations I included were either statistically significant, or if they weren’t, there was an apparent trend in the expression that was confirmed by at least one statistically significant value. This post will be pretty long and hard to summarize, which I apologize for in advance. All the data in this post is from the liver samples, I haven’t yet looked at the adipose tissue samples. mRNA was quantifiable in 100% of samples unless indicated otherwise.

BMI 22.5–25

  • Damage apparently already occurs in this BMI range
  • Receptor for retinol uptake STRA6 (STRA6) mostly turned off (-39%; p < 0.05; quantifiable in 22.7% of samples vs. 61.5% in control), indicating the presence of excess retinol and/or saturation of hepatic stellate cells (HSCs)
  • Peroxisome proliferator-activated receptor delta (PPARD) up (+47%; p < 0.01) indicating that the body is attempting to stop the weight gain by increasing beta-oxidation
  • Lecithin retinol acyltransferase (LRAT) begins increasing even though STRA6 is down, indicating that the ratio of unbound retinol to retinol esters is up, and esterification is therefore increased
  • Apparent hypoxia, evident by upregulation of hypoxia-inducible factor 1-alpha (HIF1A) (+45%; p < 0.01) without concurrent similar increases of HIF1A inhibitor (HIF1AN) or other relevant hydroxylases
  • NADH dehydrogenase [ubiquinone] 1 alpha subcomplex subunit 4-like 2 (NDUFA4L2), which oxidizes NADH begins decreasing, even though it should be induced by HIF1A (Tello, Balsa et al. 2011), indicating that even though hypoxia is apparently present, the body does not decrease its oxygen consumption in response

BMI 25–27.5

  • STRA6 off and down (-82%; p < 0.05; quantifiable in 17.9% of samples)
  • Retinol-binding protein 1 (RBP1) down (-28%; p < 0.05), indicating that the liver is potentially not only saturated with retinol but also with fat, and that the liver is attempting to prevent further esterification of unbound retinol
  • PPARD up (+12%; p < 0.05) less than in the previous range, indicating that beta oxidation is being prevented
  • Cytochrome P450 1A1 (CYP1A1) at its highest level overall (+39%; p < 0.05697), possibly indicating that retinal is no longer converted to retinoic acid, since CYP1A1 metabolizes RA and CYP enzymes have been reported to auto-regulate based on substrate availability (Morel, Mermod et al. 1999)
  • Retinal dehydrogenase 1 (ALDH1A1) up (+1%; p < 0.05) less than in the previous range (+12%; p < 0.01), possibly confirming that less retinal is oxidized to retinoic acid, since ALDH1A1 activity has been reported to be associated with phosphorylation by aurora kinase A (AURKA) activity and not substrate availability (Wang, Nikhil et al. 2017)
  • However, Aldo-keto reductase family 1 member B10 (AKR1B10) begins to get turned off (quantifiable in 59.1% of samples vs. 92.3% in control and 90.9% in the previous range), indicating that less retinal is reduced to retinol, and it is therefore unclear whether less or more retinoic acid is produced overall; no data was available for AKR1B15 which likely also plays a role
  • Retinol dehydrogenase 16 (RDH16), which prefers all-trans-retinal over 9-cis-retinal, is at its highest level overall (+23%; p < 0.05); it’s unclear whether RDH enzymes auto-regulate, this could also be a result of NAD+ depletion

BMI 27.5–30

  • AKR1B10 expression appears to increase again, however this is likely the result of an increased NAFLD to Healthy ratio compared to the previous range
  • PPARD up again (+63%; p < 0.05), indicating that beta oxidation is no longer being prevented
  • However, 17-beta-hydroxysteroid dehydrogenase (HSD17B) enzymes that prefer NADH are down, possibly indicating that fatty acids but not NADH are oxidized, or that NAD+ is diverted, and possibly indicating that the ultimate source of the problem is decreased oxygen availability
  • LRAT begins decreasing, indicating that there are now more retinyl esters, and less unbound retinol in the liver
  • CYP1A1 begins decreasing, indicating that there is now an increased amount of retinoic acid in the liver

BMI 30–35

  • CYP26A1 down (-3%; p < 0.05) compared to the previous range (+161%; not significant), indicating that there is an oversupply of either all-trans-RA, (4S)-OH-RA or 4-oxo-RA, which are all substrates of CYP26A1
  • The BMI 33–36 range is the range with the greatest NAFLD to Healthy ratio (except BMI > 48)

BMI 35–40

  • Carnitine O-palmitoyltransferase 1, liver isoform (CPT1A) at its lowest level overall (-28%; p < 0.05), indicating that beta oxidation is severely restricted, this is apparently protective since the NAFLD to Healthy ratio is decreased from 3 in the previous range to 1
  • PPAR alpha (PPARA) begins increasing (not significant), and since CPT1A is down, this could indicate that the body is transitioning from mitochondrial beta oxidation to peroxisomal beta oxidation; Peroxisomal carnitine O-octanoyltransferase (CROT), which transports lipids into peroxisomes consistently increases with BMI but the expression was not significant compared to control
  • ALDH1A1 was down compared to the previous range but this was not statistically significant

Further observations

Hepatic expression of STRA6 decreases rapidly with weight gain. STRA6 mRNA was quantifiable in 61.5% of healthy samples with BMI < 22.5 but in only 36% of healthy samples with BMI < 25. STRA6 was significantly upregulated in the BMI 20–25 group (p < 0.01) but significantly downregulated in the 25–30, 35–40 and 50–60 groups, for the most part due to not being quantifiable in more samples.

AKR1B10 is an accurate predictor of NAFLD up to BMI 36

Unexpectedly, AKR1B10 distribution was similar to STRA6 distribution. AKR1B10 mRNA was quantifiable in 92.3% of healthy samples with BMI < 22.5 but in only 68% of healthy samples with BMI < 25. Expression of AKR1B10 begins to increase at BMI 27.5–30, and STRA6 was not quantifiable in any sample from this group. Overall expression of AKR1B10 was highly variable, and it was downregulated significantly in the BMI 35–50 groups (-75%; p < 0.001). In the BMI 30–35 group AKR1B10 was +180% compared to control but this was not statistically significant due to the small number of quantifiable samples. The high variance was confirmed to be due to increased transitions to NAFLD. AKR1B10 proved to be an incredibly accurate predictor of NAFLD up to BMI 36, being quantifiable virtually only in the NAFLD/NASH samples but not in the healthy samples within the same BMI range. For BMI > 40 AKR1B10 was significantly downregulated or not quantifiable, even though NAFLD/NASH outnumbered healthy starting at BMI 45, and all samples in the BMI >50 groups were from NAFLD or NASH patients. Overall this suggests that AKR1B10 is a reliable biomarker for NAFLD up to BMI 36, however it could produce false negative results for BMI >36.

Expression of STRA6, AKR1B10, RBP1, PPARD, JAK2 and STAT3 was apparently correlated

RBP1 was down (-20%; p < 0.05) at BMI 25–30 and BMI 25–27.5 (-30%; p < 0.05), but much less at greater BMIs. This was confirmed by downregulation of just 4% at BMI 30–35 (p < 0.05). Downregulation of RBP1 apparently correlated with downregulation of both AKR1B10 and PPARD. Expression of PPARD differed significantly from the other groups but not control (p < 0.05) at BMI 20–25 and 30–35 (p < 0.01), and differed significantly from control (+3%; p < 0.05) at BMI 25–30. Further separation revealed that expression of PPARD was up by roughly 40% compared to control in the BMI 20–22.5 (p < 0.05), 22.5–25 (p < 0.01), 27.5–30 (p < 0.01) and 30–32.5 groups (p < 0.01), however in the BMI 25–27.5 group PPARD was only up 12% (p < 0.05 vs control and p < 0.001 vs other groups). Overall there was an apparent correlation between STRA6, AKR1B10, RBP1, PPARD and tyrosine-protein kinase JAK2 (JAK2). JAK2 was significantly downregulated at BMI 25–30, coinciding with decreased expression of PPARD. Additionally, there was no apparent correlation between signal transducer and activator of transcription 5A (STAT5A) or STAT5B and JAK2/PPARD, there was however a correlation between STAT3 and the other genes. STAT3 was significantly upregulated (+21%; p < 0.05) at BMI 22.5–25, coinciding with upregulation (+47%; p < 0.01) of PPARD. At BMI 25–30 however STAT3 was downregulated (-13%; p < 0.01), coinciding with decreased expression of RBP1 (-20%; p < 0.05) and JAK2 (-15%; p < 0.01). These results suggest that on activation by STRA6, JAK2 primarily phosphorylates STAT3 and not STAT5.

Retinoic acid decreases transcriptional activity of STAT3 independent of SOCS3

Protein kinase C epsilon type (PRKCE) is up in the BMI 32.5–35 range (+117%; p < 0.05) compared to the previous range (+13%; p < 0.05), and this upregulation coincided with downregulation of SOCS3 (-22%; p < 0.05) compared to the previous range (+346%; p < 0.01), while STAT3 expression was unchanged. Since SOCS3 is known to be a feedback inhibitor of STAT3 (Gao, Zhao et al. 2018), and additionally PRKCE phosphorylates STAT3 (UniProt), this seems to indicate (but does not prove) that STAT3 induces transcription of PRKCE, which phosphorylates STAT3, leading to increased transcription of PRKCE, and this cycle repeats until STAT3 is inhibited by SOCS3. Furthermore, retinoic acid (RA) and holo-RBP1 have been reported to competitively inhibit the phosphorylytic activity of PRKCA, and that PRKCE displayed a higher affinity for atRA than PRKCA (Radominska-Pandya, Chen et al. 2000). Inactivation of PRKCE with siRNA was shown to reduce cellular P-STAT3 Tyr but not P-STAT3 Ser (Shi, Papay et al. 2012). It has been shown that Tyr705 phosphorylation of STAT3 induces dimerization and translocation of STAT3 to the nucleus (Guadagnin, Narola et al. 2015), whereas phosphorylation of Ser727 increases transcriptional efficiency but induces localization to mitochondria instead of nucleus. Therefore  retinoic acid, by inhibiting PRKCE, apparently inhibits formation of P-STAT3 Tyr but not P-STAT3 Ser, indicating that, overall, retinoic acid decreases transcriptional activity of STAT3 and promotes migration of the more potent P-STAT3 Ser to mitochondria. Additionally, this appears to confirm that increased transcription of PRKCE leads to increased transcriptional activity of STAT3 via increased Tyr705 phosphorylation.

This theory is further supported by the finding that expression of the STAT3 target genes HIF1A (Carpenter and Lo 2014), NNMT and LSM10 (Kulesza, Ramji et al. 2019) apparently correlated with expression of STAT3 in the BMI 22.5–25, 25–27.5 and 27.5–30 ranges, but then diverged in the BMI 32.5–35 range, which could be explained by decreased phosphorylation of STAT3 by PRKCE due to inactivation of PRKCE. However, none of HIF1A, LSM10, NNMT and STAT3 differed significantly from control in the BMI 32.5–35 range, which could be a result of varying concentrations of retinoic acid in the BMI 32.5–35 samples. The increased presence of retinoic acid is confirmed by downregulation of CYP1A1 and CYP26A1, and further confirmed by downregulation of PTGS2 (-30%; p < 0.05) compared to the previous range (-0%; not significant), which has been reported to be both required for the metabolism of retinoic acid (Samokyszyn, Freyaldenhoven et al. 1997) and downregulated by retinoic acid (Han, Zhang et al. 2014).

To summarize, expression of the known STAT3 target genes HIF1A, NNMT, LSM10 roughly correlated with expression of STAT3 from BMI 25–32.5, but not in the BMI 32.5–35 group, a group that is known to exhibit increased cellular retinoic acid concentrations. PRKCE, which has not been reported to be a STAT3 target gene, but which induces transcription of genes by STAT3 showed a similar expression pattern. In the BMI 32.5–35 group, STAT3 target genes should have been overexpressed due to decreased expression of SOCS3 and increased expression of PRKCE, but that was found not to be the case. This divergence could be explained by increased inactivation of PRKCE by retinoic acid, resulting in a SOCS3-independent decrease of STAT3 transcriptional activity. Undesired inactivation of PRKCE by retinoic acid would likely also lead to increased transcription of PRKCE as a compensatory mechanism.

Interestingly, racemic 4-OH-RA was shown inhibit PRKCA activity roughly 50% less than atRA (Radominska-Pandya, Chen et al. 2000). It should be investigated whether (4S)-OH-RA and (4R)-OH-RA differentially inhibit PRKCA and the remaining protein kinases C.

ISX signaling promotes transcription of wound-sensing proteins

Quote from Jenny on October 11, 2021, 4:45 am

‘ISX suppresses gene expression of the scavenger receptor class B type 1 (Scarb1) and the β-carotene-15,15′-dioxygenase (Bco1), which encode proteins that respectively mediate the uptake of carotenoids and their conversion into retinoids (810). This negative feedback regulation controls the utilization of dietary BC for retinoid production in mice depending on demand and availability (9, 10).’ 

(from paper linked in above comment) 

If ISX expression is compromised it can be seen how things go awry. 

As Jenny noted, intestine-specific homeobox (ISX) has been reported to repress transcription of beta,beta-carotene 15,15’-dioxygenase (BCO1), and is apparently inducible by retinoic acid. ISX was up (+188%; p < 0.01; quantifiable in 78% of samples vs 48% in control) in the BMI 25–30 group, variable in the BMI 30–40 groups and decreasing thereafter. ISX has also been reported to directly induce transcription of transcription factor E2F1 (E2F1) by binding to the E2F1 promoter (Wang, Wang et al. 2016). Even though in my analysis E2F1 did not differ significantly from control in any group, a somewhat weak correlation between ISX and E2F1 is apparent until about BMI 35 in both gene expression and quantifiability. Interestingly, E2F1 in complex with retinoblastoma-associated protein (RB1) was reported to repress transcription of genes related to oxidative metabolism and mitochondrial biogenesis in both muscle and adipocytes (Denechaud, Fajas et al. 2017). Furthermore, E2F1 was shown to be required for transcription of discoidin domain receptor family, member 1 (DDR1), which is activated by collagen and capable of forming both P-STAT3 Tyr and P-STAT3 Ser (Wang, Sun et al. 2017). Conversely, STAT3 also induces expression of DDR1 (Lin, Jin et al. 2020). This regulatory mechanism appears to be very similar to that of PRKCE. Surprisingly, my results show that expression of E2F1 was highly correlated with expression of PRKCE.

It has also been shown that when STAT3 is activated in a cell type-independent mechanism, it is recruited to DNA sites with E2F1 already pre-bound before STAT3 activation (Hutchins, Diez et al. 2013), which could explain the correlation between PRKCE and E2F1. Conceptually, it appears that both PRKCE and DDR1 are responsible for detecting different types of wounds. PRKCE is activated by diglycerides (DAG) produced during hydrolysis of phospholipids, and also by ROS produced during hypoxia-induced mitochondrial dysfunction (Wikipedia), whereas DDR1 is activated by free collagen, which can also be indicative of wounds. Collagen Type I was found to be capable of inducing transformation of epithelial cells to malignant mesenchymal stem cells, and additionally it has been confirmed that HSCs secrete collagen I (Yang, Wang et al. 2014)

It appears that 13-cis-retinoic acid first hydrolyzes phosphatidylinositol 4,5-bisphosphate during photochemical isomerization to atRA, producing DAG (Tang and Ziboh 1991), and subsequently atRA deactivates the PRKCs responsible for sensing this damage. Activation of PRKCs by the nonspecific activator 12-O-tetradecanoylophorbol-13-acetate (TPA) was shown to enhance the tight junction barrier function of human nasal epithelial cells (Koizumi, Kojima et al. 2008).

Therefore, it seems like activation of ISX is primarily protective, since it induces transcription of E2F1, which is required for transcription of PRKCE and DDR1 by STAT3. Since expression of PRKCE highly correlated with expression of E2F1, this would suggest that transcription of E2F1, and not activation of STAT3 could be the rate limiting factor. Interestingly, E2F1 was downregulated (-38%; p = 0.071) in BCO1-null mice (Smith, Ford et al. 2016) which, philosophically, could indicate that a major function of E2F1 is to limit the damage caused by dietary carotenoids. 


Even though expression of PPAR gamma (PPARG) consistently increased in every BMI range, this effect was not statistically significant except in the BMI > 60 group (+13%; p < 0.01). PPARA was downregulated in the BMI 25–30 group (-8%; p < 0.05), upregulated in the BMI 40–50 group (+7%; p < 0.05) and downregulated in the BMI 50–60 (-12%; p < 0.05) and >60 (-10%; p < 0.01) groups. It is possibly that PPARA activity is protective, since expression was increased in groups with lower NAFLD/Healthy ratios, namely BMI 36–45.

Edit: Since this post is very long, let me reiterate the most important point. Apparently retinoic acid, after damaging cell membranes, deactivates the proteins responsible for detecting this damage, and the damage is therefore not repaired.


Carpenter, R. L. and H.-W. Lo (2014). "STAT3 Target Genes Relevant to Human Cancers." Cancers 6(2): 897-925.

Denechaud, P.-D., L. Fajas and A. Giralt (2017). "E2F1, a Novel Regulator of Metabolism." Frontiers in Endocrinology 8(311).

Gao, Y., H. Zhao, P. Wang, J. Wang and L. Zou (2018). "The roles of SOCS3 and STAT3 in bacterial infection and inflammatory diseases." Scand J Immunol 88(6): e12727.

Guadagnin, E., J. Narola, C. G. Bönnemann and Y.-W. Chen (2015). "Tyrosine 705 Phosphorylation of STAT3 Is Associated with Phenotype Severity in TGFβ1 Transgenic Mice." BioMed research international 2015: 843743-843743.

Han, J., L. Zhang, X. Chen, B. Yang, N. Guo and Y. Fan (2014). "Effects of all-trans retinoic acid on signal pathway of cyclooxygenase-2 and Smad3 in transforming growth factor-β-stimulated glomerular mesangial cells." Exp Biol Med (Maywood) 239(3): 272-283.

Hutchins, A. P., D. Diez, Y. Takahashi, S. Ahmad, R. Jauch, M. L. Tremblay and D. Miranda-Saavedra (2013). "Distinct transcriptional regulatory modules underlie STAT3’s cell type-independent and cell type-specific functions." Nucleic Acids Research 41(4): 2155-2170.

Koizumi, J., T. Kojima, N. Ogasawara, R. Kamekura, M. Kurose, M. Go, A. Harimaya, M. Murata, M. Osanai, H. Chiba, T. Himi and N. Sawada (2008). "Protein kinase C enhances tight junction barrier function of human nasal epithelial cells in primary culture by transcriptional regulation." Mol Pharmacol 74(2): 432-442.

Kulesza, D. W., K. Ramji, M. Maleszewska, J. Mieczkowski, M. Dabrowski, S. Chouaib and B. Kaminska (2019). "Search for novel STAT3-dependent genes reveals SERPINA3 as a new STAT3 target that regulates invasion of human melanoma cells." Laboratory Investigation 99(11): 1607-1621.

Lin, Y., H. Jin, X. Wu, Z. Jian, X. Zou, J. Huang, R. Guan and X. Wei (2020). "The cross-talk between DDR1 and STAT3 promotes the development of hepatocellular carcinoma." Aging 12(14): 14391-14405.

Morel, Y., N. Mermod and R. Barouki (1999). "An autoregulatory loop controlling CYP1A1 gene expression: role of H(2)O(2) and NFI." Mol Cell Biol 19(10): 6825-6832.

Radominska-Pandya, A., G. Chen, P. J. Czernik, J. M. Little, V. M. Samokyszyn, C. A. Carter and G. Nowak (2000). "Direct interaction of all-trans-retinoic acid with protein kinase C (PKC). Implications for PKC signaling and cancer therapy." J Biol Chem 275(29): 22324-22330.

Samokyszyn, V. M., M. A. Freyaldenhoven, H. C. Chang, J. P. Freeman and R. L. Compadre (1997). "Regiospecificity of peroxyl radical addition to (E)-retinoic acid." Chem Res Toxicol 10(7): 795-801.

Shi, T., R. S. Papay and D. M. Perez (2012). "α(1A)-adrenergic receptor differentially regulates STAT3 phosphorylation through PKCϵ and PKCδ in myocytes." Journal of receptor and signal transduction research 32(2): 76-86.

Smith, J. W., N. A. Ford, J. M. Thomas-Ahner, N. E. Moran, E. C. Bolton, M. A. Wallig, S. K. Clinton and J. W. Erdman, Jr. (2016). "Mice lacking β-carotene-15,15'-dioxygenase exhibit reduced serum testosterone, prostatic androgen receptor signaling, and prostatic cellular proliferation." American journal of physiology. Regulatory, integrative and comparative physiology 311(6): R1135-R1148.

Tang, W. and V. A. Ziboh (1991). "Phorbol ester inhibits 13-cis-retinoic acid-induced hydrolysis of phosphatidylinositol 4,5-bisphosphate in cultured murine keratinocytes: a possible negative feedback via protein kinase C-activation." Cell Biochem Funct 9(3): 183-191.

Tello, D., E. Balsa, B. Acosta-Iborra, E. Fuertes-Yebra, A. Elorza, Á. Ordóñez, M. Corral-Escariz, I. Soro, E. López-Bernardo, E. Perales-Clemente, A. Martínez-Ruiz, J. A. Enríquez, J. Aragonés, S. Cadenas and M. O. Landázuri (2011). "Induction of the mitochondrial NDUFA4L2 protein by HIF-1α decreases oxygen consumption by inhibiting Complex I activity." Cell Metab 14(6): 768-779.

Wang, J., K. Nikhil, K. Viccaro, L. Chang, J. White and K. Shah (2017). "Phosphorylation-dependent regulation of ALDH1A1 by Aurora kinase A: insights on their synergistic relationship in pancreatic cancer." BMC Biology 15(1): 10.

Wang, S.-N., L.-T. Wang, D.-P. Sun, C.-Y. Chai, E. Hsi, H.-T. Kuo, K. K. Yokoyama and S.-H. Hsu (2016). "Intestine-specific homeobox (ISX) upregulates E2F1 expression and related oncogenic activities in HCC." Oncotarget 7(24): 36924-36939.

Wang, Z., X. Sun, Y. Bao, J. Mo, H. Du, J. Hu and X. Zhang (2017). "E2F1 silencing inhibits migration and invasion of osteosarcoma cells via regulating DDR1 expression." International journal of oncology 51(6): 1639-1650.

Yang, M.-C., C.-J. Wang, P.-C. Liao, C.-J. Yen and Y.-S. Shan (2014). "Hepatic stellate cells secretes type I collagen to trigger epithelial mesenchymal transition of hepatoma cells." American journal of cancer research 4(6): 751-763.

Jenny, Beata and 4 other users have reacted to this post.