Evidence for a Causal Association Between Human [CMV] Infection and Chronic Back Pain: A One‐Sample Mendelian Randomization Study, 2025, Naeini et al

[forestglip]

Evidence for a Causal Association Between Human Cytomegalovirus Infection and Chronic Back Pain: A One‐Sample Mendelian Randomization Study

Maryam Kazemi Naeini, Maxim B Freidin, Isabelle Granville Smith, Stephen Ward, Frances M K Williams

Background
Chronic back pain (CBP) is a major cause of disability globally. While its etiology is multifactorial, specific contributing genetic and environmental factors remain to be discovered. Paraspinal muscle fat has been shown in human and preclinical studies to be related to CBP. One potential risk factor is infection by cytomegalovirus (CMV) because CMV is trophic for fat. CMV may reside in the paraspinal muscle adipose tissue. We set out to test the hypothesis that previous CMV infection is linked to CPB using a one‐sample Mendelian randomization (MR).

Method
The sample comprised 5140 UK Biobank participants with information about CMV serology and CBP status. A one‐sample MR based on independent genetic variants predicting CMV positivity was conducted in Northern European participants. To validate the association further, the MR study was repeated using a CMV polygenic risk score (PRS). As a negative control for confounding and spurious causal inference, we used Epstein–Barr virus (EBV) serology, because EBV is another common viral infection but is not trophic for adipose tissue.

Results
A genome‐wide association study for CMV seropositivity revealed 86 independent SNPs having p‐value < 2×10−4 that have been used to define genetically‐predicted categories of CMV infection risk. The CMV predicted categories were found statistically significantly associated with CBP (OR = 1.150; 95% CI: 1.005–1.317, p‐value = 0.043). Stronger significant results were obtained using the PRS for CMV seropositivity (OR = 1.290; 95% CI: 1.133–1.469, p‐value = 12E‐4). No such association was seen between EBV and CBP.

Conclusion
Our results provide evidence for a causal relationship between CMV infection and CBP. Further investigation is warranted to get insight into the mechanism by which CMV might contribute to the pathogenesis of CBP.

Link | PDF (Spine) [Open Access]

 

[Jonathan Edwards]

I have only looked at the abstract but I am not sure how they conclude that there is a causal relationship, unless they mean that a certain gene make-up seems to confer risk for both CMV infection and back pain.

It sounds a bit odd to blame back pain on paraspinal fat infection since the pain localises to a well documented area of disc deterioration.

Maybe I have not understood how the causal path is supposed to work.

 

[Utsikt]

It sounds a bit odd to blame back pain on paraspinal fat infection since the pain localises to a well documented area of disc deterioration.

I’ve often heard that people without back pain also often have things ‘wrong’ with their spine - and that that means that the ‘damage’ likely isn’t the cause of the pain.

Would you say that’s wrong?

 

[forestglip]

I have only looked at the abstract but I am not sure how they conclude that there is a causal relationship, unless they mean that a certain gene make-up seems to confer risk for both CMV infection and back pain.

Yeah, I'm still trying to wrap my head around MR but I don't see how this shows CMV is causal for CBP as opposed to the reverse or the gene being causal for both.

It sounds a bit odd to blame back pain on paraspinal fat infection since the pain localises to a well documented area of disc deterioration.

Maybe I have not understood how the causal path is supposed to work.

They speculate a bit here. It seems basically they think it's possible CMV infects adipose tissue in the back or intervertebral discs.

There are several hypothetical ways by which CMV may cause CBP.

One of them is the impact of CMV on intervertebral disc degeneration (IVDD) which is the major risk factor for CBP, with individuals with the highest IVDD having a threefold greater risk. Alpantaki et al. detected high CMV infection incidence in intervertebral disc and identified herpes virus DNA in the intervertebral disc of individuals with lumbar disc herniation, suggesting that herpes viruses may play a role in the etiology of degenerative disc disease [11].

Another link between CMV and CBP includes t adipose tissue in and around the muscles that support the spine and are known to contribute to CBP. In CBP, the so-called paraspinal muscles exhibit increased adipose deposition and atrophy [12]. Unilateral back pain is associated with paraspinal fat infiltration on the same side as the symptoms, and animal studies support the concept that pain is produced by muscle fat infiltration [13, 14]. Human adipose tissue stimulates active CMV infection, and CMV may target adipocytes. Adipose tissue infection may play an important role in the etiology of CMV condition [15]. Contreras et al. showed that human white adipose tissues as a large reservoir of CMV-specific T lymphocytes, most likely used to manage latent viral reactivation [9].

 

[Jonathan Edwards]

I’ve often heard that people without back pain also often have things ‘wrong’ with their spine - and that that means that the ‘damage’ likely isn’t the cause of the pain.

Would you say that’s wrong?

I think that is misleading. You might say that a lot of people have tiles missing on the roof who don't have water coming in. That doesn't mean that some people don't have water coming in because tiles are missing (as I have direct proof of).

Almost everyone has degeneration of the cartilage of the joint at the base of the big toe (bunion joint) by sixty. Only a small proportion have pain. But it is probably the commonest joint in the foot to hurt as you get older. Equally, the joint gets a bit thicker in almost everyone. In a small proportion it rubs on a shoe and develops a 'bunion' bursa. But the fact that a lot of people don't have the bunion bursa doesn't mean it is due to the thickened joint rubbing.

To me the argument quoted just doesn't follow.

 

[Jonathan Edwards]

Yeah, I'm still trying to wrap my head around MR but I don't see how this shows CMV is causal for CBP as opposed to the reverse or the gene being causal for both.

I read your summary, which is very helpful.

I think maybe they have good reason to think that some genes create risk for a process that contributes both to CMV infection and back pain. In a sense that is a causal relation but an epiphenomenal 'steam whistle' one.

Ask a back pain sufferer I have noted that whenever I have any viral infection my back pain (due to severe degeneration of the L4-5 disc) is worse and sometimes worse enough to stop me walking upright. In between times there are times when I forget I have aback problem. My guess is that during viral infection there are systemic changes to nerve sensitivity mediated either by cytokines or purely neural signals possibly. It might be that CMV tricks the immune system (every common organism tricks the immune system one way or another) in some way that depends on your settings for cytokine production. The same settings may mean you have more nerve sensitisation with any old virus.

 

[wigglethemouse]

You need to be careful how you use the genetic database. For example there are a number of genes associated with the number of cars in a household and certain variants that have p < 10-8.
Just a completely random example.
http://geneatlas.roslin.ed.ac.uk/trait/?traits=662

 

[Jonathan Edwards]

For example there are a number of genes associated with the number of cars in a household

Yes, that's our next-door neighbours' pedigree.
Do you have a source for that @wigglethemouse? It was something that worried me when DecodeME was being designed and I hassled Chris Ponting about it - genes that encode for answering internet requests for patient DNA for instance (different alleles encoding for answering internet requests for healthy DNA).

I have been impressed how well they have handled those sorts of issues in Edinburgh. King's ought to know how to handle them but you never can be sure.

 

[wigglethemouse]

Do you have a source for that @wigglethemouse?

Yes, it is the UK Biobank dataset search tool hosted by the University of Edinburgh (well it has there logo).
http://geneatlas.roslin.ed.ac.uk/trait/?traits=662

 

[forestglip]

Yes, it is the UK Biobank dataset search tool
http://geneatlas.roslin.ed.ac.uk/trait/?traits=662

Do you know if there's any way for the general public to see which specific genotypes are significant for traits on there?

 

[wigglethemouse]

Do you know if there's any way for the general public to see which specific genotypes are significant for traits on there?

Search=>Region of gene.
e.g. Numbers of cars in household for region 6 shown as a plot (another option is table)
http://geneatlas.roslin.ed.ac.uk/re...10000&minregion=0&chrom=6&representation=plot

There is a way to get access to the raw data on the UK Biobank website which a user here did to disprove the genetic association Chris Ponting (or colleague) found for ME/CFS - that case showed the variant was for one person and was very rare - something that should not be included with proper QA - for example dismissing variants with an allele frequency below a certain cutoff.

EDIT : To clarify/correct - the mistake it was in a Chris Ponting guest Blog identifying P4HA1.

Chris admitted the error and the gene was listed in the final paper but shown as not relevant due to a MAF of 0.00029 despite a p of 2x10E-12. Paulo Maccallini was the one that found the error and Chris acknowledged the error here. Explanation of how the error was found was in the comments of either that blog, or Paulo's blog, but relevant blog pages no longer have comments showing. I seem to remember that it was the comments that provided a link to the raw data.

 

[forestglip]

Search=>Region of gene.
e.g. Numbers of cars in household for region 6 shown as a plot (another option is table)
http://geneatlas.roslin.ed.ac.uk/re...10000&minregion=0&chrom=6&representation=plot

Thanks for that. I looked around and found that Search > By Significance was more what I wanted. Just a list of the SNPs. I don't really understand what the imputed ones are so I clicked to sort by "imp. score" twice and the 6 genotyped SNPs for vehicle count are listed first.

One SNP is related to alcohol metabolism which is kind of interesting.

 

[jnmaciuch]

From my understanding (remembering one lecture several months ago), the idea of MR is to find alleles that are so strongly correlated to your “risk exposure” that they are basically a proxy for it.

For example, if you wanted to know whether there’s a causal relation between a certain protein level with some health outcome, you might zero in on an allele with a premature stop codon that renders the protein inert. In that way, it’s like doing an experiment where you managed to knock down the protein in the person.

Since an allele present from birth is very unlikely to be affected by environmental/lifestyle factors, you’re minimizing confounding variables as much as you possibly can in an observational study—making it approach the level of control of confounders you’d have in an empirical setup (which is what allows causal inference).

So for this study, they basically zeroed in on alleles that are so highly correlated with CMV susceptibility that having them basically guaranteed CMV infection at some point in life. Added: Then they check how much their “guarantee” for CMV infection is also associated with back pain.

The ability to call it a causal relationship depends on how much their “proxy” genes fulfill a set of assumptions, including their degree of proxy-ness.

The use of EBV as negative control also further accounts for confounders that would originate just from having a viral infection. The idea being that if any similar infection confers the same risk for back pain, you’d see it come up with confirmed EBV infection as well. So by filtering out the ones that were also associated with EBV, they can be more certain that the relationship is CMV-specific.

Hopefully I’m getting the explanation right here—it’s been a while since I learned about it. They specifically noted that they had to raise their p-value threshold in this study to find their strongly-related-alleles, which they seem to attribute to smaller sample size. I don’t think I have the expertise to determine if they overcame that limitation successfully, though.

 

[forestglip]

So for this study, they basically zeroed in on alleles that are so highly correlated with CMV susceptibility that having them basically guaranteed CMV infection at some point in life.

This part I'm not sure is right. I don't think the initial alleles necessarily have to be nearly perfectly associated with the variable in question. At least in this paper, it seems like they attempted to just pick out any alleles that match the standard GWAS significance threshold (and then chose a lower threshold when they didn't find enough alleles).

To reveal genetic variants for Mendelian randomization (MR), we carried out a genome-wide association study (GWS) for CMV and EBV seropositivity using the assayed sample of 8808 Northern Europeans having CMV and EBV seropositivity results. The GWAS was performed with imputed variants using GCTA, adjusted for sex and age. The criteria to filter variants were MAF > 0.005, imputation INFO score > 0.7, and missingness rate < 0.02.

However, the limited sample size precluded the identification of a sufficient number of independent SNPs at the conventional genome-wide significance threshold.

Therefore, we used a more liberal p-value threshold of p < 5 × 10−4 to capture additional potentially informative variants. Although this threshold may include variants with weaker evidence, we mitigated this concern by conducting a one-sample MR analysis using individual-level data, which enhances statistical power under these constraints.

And then for example there are MR studies looking for whether something like depression causes some other condition. I think it'd be pretty big news if there were SNPs that basically guaranteed getting depression.

 

[forestglip]

I'm only just learning the basics, but I have questions about this study. I'm looking at a paper about Mendelian randomization that makes me think they might not have fulfilled the required assumptions:

The first assumption is basically just finding alleles that are associated with CMV in the GWAS. They lowered the threshold for inclusion quite a bit from the standard GWAS threshold (p<5x10⁴ instead of p<5x10⁸), so I'm not sure how associated they really are.

Relevance Assumption: The Genetic Variant Must Be Robustly Associated with the Exposure

The most common method of deriving genetic instruments in recent MR studies is via GWAS, whereby single-nucleotide polymorphisms (SNPs) that pass genome-wide significance (p < 5 × 10−8) are typically considered for inclusion. However, it is important that the strength of the instrument is tested separately to appraise the relevance assumption, which is often done by means of the proportion of variance explained (r2) and the related F-statistic, which additionally takes into account the size of the sample under investigation. Increasingly, multiple genetic variants are found to be independently associated with traits investigated in GWAS and these may be combined in genetic risk scores or through meta-analysis approaches to explain more variation in the trait (Dudbridge 2020). This in turn can be used to increase power, obtain more precise causal estimates and minimize risk of weak instrument bias (i.e., uncertainty in the SNP-exposure association that can bias causal estimates) (Pierce and Burgess 2013).

The second assumption is that there are no common causes of both the alleles in question and back pain. For example if people from Europe happen to more often have these alleles, and people from Europe also happen to more often have back pain.

Independence/Exchangeability Assumption: There Are No Confounders of the Association between the Genetic Variant and Outcome

[...] Concerns about potential violation of this assumption at a population level relate to confounding by ancestry or population stratification, which can influence variation in both allele frequency and disease risk in population(s) being investigated (Fig. 3). Approaches to limit spurious associations generated because of population groups include use of genetic associations derived from homogeneous populations or with adequate control for population structure (e.g., through principal components analysis or linear mixed models) (Loh et al. 2015). However, the independence assumption can also be violated by dynastic effects (when parental genotypes directly affect offspring phenotypes), or by assortative mating (when individuals select a partner based on a particular phenotype). These biases will likely differ depending on the exposure(s), outcome(s), and population(s) under study.

It is impossible to prove the independence assumption in an MR study because, although attempts can be made to account for ancestry and examine how genetic variants relate to measured confounders, associations with unknown confounders cannot be demonstrated. In addition, whereas previous recommendations have been to assess associations between the genetic instrument and a wide range of factors that could bias exposure–outcome associations (Davey Smith et al. 2007), these associations are likely to be indicators of confounding by ancestry (Fig. 3) or horizontal pleiotropy (Fig. 4), rather than reflecting conventional confounding.

The third assumption is that the alleles do not cause the outcome (back pain) independently of the risk factor (CMV infection). For example if an allele both suppressed the immune system to increase risk of CMV infection and also directly affected the structure of a protein in the spine that led to pain, then you couldn't say that the study shows that CMV infection increases risk of back pain, since it might just be the gene directly increasing the risk of both. This would be called horizontal pleiotropy, and is the main thing I'm wondering about with this study. How do we know these alleles don't cause back pain independently of CMV?

Exclusion Restriction Assumption: The Genetic Variant Should Only Influence the Outcome of Interest via the Exposure
Pleiotropy is the phenomenon whereby a genetic variant influences multiple traits, and is a major threat to the exclusion restriction assumption. However, it is important to make the distinction between vertical and horizontal pleiotropy (Davey Smith and Hemani 2014; Hemani et al. 2018a). Vertical (or mediated) pleiotropy occurs when the genetic variant (G) is associated with the outcome (Y) because G affects Y entirely through the exposure (X). This fulfils the exclusion restriction assumption and is the essence of the MR approach. Horizontal (unmediated or biological) pleiotropy occurs when G affects both X and Y but through different pathways. This can yield biased estimates in MR if a genetic instrument influences the outcome via a mechanism other than the exposure of interest (Verbanck et al. 2018). Such pleiotropy can be direct, as in the path from G to Y (uncorrelated pleiotropy), or can be indirect (e.g., when G affects X and Y through a shared confounder, U [correlated pleiotropy]) (Fig. 4; Morrison et al. 2020). The latter may occur in cases of misspecifying the primary phenotype, such as when a genetic variant is used to proxy for a trait secondary to the trait with which it is directly associated.

Although it is not possible to prove that the exclusion restriction assumption holds in any MR study, various sensitivity analyses can be applied to uncover deviations from the assumption.

They give an example of a method to test the last assumption:

In some instances, conducting a stratified analysis can provide evidence against the possibility of horizontal pleiotropy. When a genetic variant is not related to the exposure of interest in a particular subgroup of the population, this variant should also not be associated with the outcome of interest in this subgroup (given an absence of the association with the exposure). For example, ALDH2, coding for aldehyde dehydrogenase 2, is a common polymorphism in East Asian populations that has been used as a genetic instrument for alcohol consumption (Lewis and Davey Smith 2005; Chen et al. 2008; Millwood et al. 2019). In East Asian populations, in which women are much less likely to drink alcohol than men, this polymorphism is not strongly associated with alcohol intake among women (Chen et al. 2008). This approach has been used to assess the presence of pleiotropy and evaluate a causal relationship between alcohol consumption and increased blood pressure (Chen et al. 2008) and risk of vascular disease (Millwood et al. 2019). For example, if the effects of alcohol consumption on blood pressure and vascular disease are causal, we would expect to find evidence of association between variation in ALDH2 and the outcomes in East Asian men, but not East Asian women. Any association observed between ALDH2 and the outcomes in East Asian women, in the absence of alcohol intake, would indicate pleiotropy. Such an approach can be considered a negative control design (Lipsitch et al. 2010; Davey Smith 2012b) and models built on this approach can detect and adjust for the pleiotropic effects and provide valid estimates in such instances (Cho et al. 2015; Spiller et al. 2019) (see the section “Methods for Assessing and Accounting for Horizontal Pleiotropy”). However, genetic variants that are not associated with the exposure in a subgroup of a population may be uncommon, and so such direct assessment of pleiotropy is often not possible.

This would be doing something like only looking at the subset of people who are seronegative for CMV. If the alleles in question are still associated with back pain in this subset of people, then it's more likely that the main hypothesis - allele causes increased CMV which causes back pain - is not correct, since the alleles are associated with back pain in the absence of CMV. I don't see that they did something like this.


They did do some other kind of control, where they also looked for alleles associated with EBV seropositivity, then checked if those alleles were associated with back pain as well, which they were not. I guess this was to check that it is CMV specifically, and not infections in general, that is the risk factor for back pain.

EBV measured concurrently with CMV in the UK Biobank—was employed as a negative control. This approach allowed us to assess potential confounding by other viral exposures and to ensure the robustness of the observed association between CMV seropositivity and chronic back pain.

They say they used some tools to do the MR. Maybe these take care of doing tests to check that these assumptions hold?

In a one-sample MR study, we used instrumental variable analysis with two-stage least-squares regression to investigate the causal relationship between genetically determined CMV and EBV using the ivtools package in R.

Anyways, I'm wondering if this study is rigorous enough to say this:

Our results provide evidence for a causal relationship between CMV infection and CBP.

the observed causal association between CMV and CBP is most likely genuine.

I know very little about MR, so please no one take my ramblings as meaningful. They may very well have done everything the right way, and I just don't understand it.

 

[jnmaciuch]

This part I'm not sure is right. I don't think the initial alleles necessarily have to be nearly perfectly associated with the variable in question. At least in this paper, it seems like they attempted to just pick out any alleles that match the standard GWAS significance threshold (and then chose a lower threshold when they didn't find enough alleles).

Yes, sorry, I was exaggerating for the purposes of brief explanation. Individually an allele is not likely to be a perfect predictor for anything other than a Mendelian disease. It’s likely to be a polygenic trait, which is why they use GWA to generate a polygenic risk score.

And then the polygenic risk score won’t be a perfect predictor numerically because one individual will likely not have every single risk variant associated with the exposure—they’ll only have a couple. But the point of this step is to find a strong enough allelic predictor score that it can serve as a reasonable proxy for CMV infection. That’s what assumption one in your linked paper is getting towards

 

[jnmaciuch]

How do we know these alleles don't cause back pain independently of CMV?

That would be part of the screening for allelic traits to include/exclude from the polygenic risk score. Alleles would be included if they were associated with back pain only when CMV infection was also confirmed (assumption 3 in your linked paper). If an allele was also associated with back pain in individuals who were seronegetive, it’s filtered out.

Added: generating an MR proxy is the process of filtering down to a list of predictors that don’t violate any of the MR assumptions. That doesn’t completely guarantee that the predictors don’t violate the assumptions in a way that the authors couldn’t think of, it just means they’ve excluding anything that clearly violates them.

They might then do additional filtering to address things that might be less obvious, which is why they’re doing the EBV negative control. So that’s a major limitation of these types of studies, but in that way it is like an experimental setup where you just might not be aware of confounders despite your best efforts to predict and mitigate them.

The main difference is that a laboratory setting can usually limit the amount of uncontrolled confounders—in a population-wide study like this, the potential for confounders is increased by virtue of variability in humans and human life.

Added: focusing on alleles in the first place is meant to limit that potential for confounders substantially for reasons I already mentioned (lifestyle factors that make you more likely to catch CMV won’t actually change someone’s alleles), but it’s not a guarantee. The hope is that the alleles are randomly distributed amongst the population to control additional confounders like you would in a randomized clinical trial.

 

[forestglip]

That would be part of the screening for allelic traits to include/exclude from the polygenic risk score. Alleles would be included if they were associated with back pain only when CMV infection was also confirmed (assumption 3 in your linked paper). If an allele was also associated with back pain in individuals who were seronegetive, it’s filtered out.

Added: generating an MR proxy is the process of filtering down to a list of predictors that don’t violate any of the MR assumptions. That doesn’t completely guarantee that the predictors don’t violate the assumptions in a way that the authors couldn’t think of, it just means they’ve excluding anything that clearly violates them.

Ah, are you saying some of that filtering is implied in an MR study and was probably done here? I was assuming there'd be explicit details about their methods for that.

I'm looking at the depression MR study I linked above. It's in a Nature journal so I assume there's a better chance everything is done properly.

I notice they also relax the threshold:

In cases where traits lacked an adequate number of SNPs, we applied a more relaxed instrument threshold of P < 1×10–5 to ensure an adequate SNP count.

But then they also say they run a test, which was mentioned in the MR guidelines I quoted above, to make sure these SNPs are not "weak instruments". The thread's paper doesn't mention doing this.

Additionally, to safeguard against any influence of weak instrumental biases on causal inference, we gauged the strength of the genetic instruments for all remaining SNPs, calculating the F statistic as β2/se2. An F statistic greater than 10 is deemed a robust instrumental variable and is considered appropriate for use in MR studies [15].

For the other assumptions, they list the tests they ran to check that. I don't know the details of any of these tests, but that's the kind of thing I was looking for in the thread's paper, which seems kind of low on detail.

To further probe potential horizontal pleiotropy, we employed the MR-Egger intercept test. The MR-PRESSO method is designed to identify SNP outliers with pleiotropic effects, and subsequently, it offers an estimate that becomes consistent with the one obtained through the IVW (Inverse Variance Weighted) method after these outliers are excluded [24]. The leave-one-out sensitivity method was used to assess the effect of individual SNPs on causality [25]. We also performed the MR-Steiger directionality test to infer the direction of causality [26].

[...]

In MR-Egger intercept test, there was no significant horizontal pleiotropy in all analyses (all På 0.05) (Table S8). Heterogeneity was nor detected in the Cochran’s Q analyses. In addition, no outlier was detected by MR-PRESSO approach (Table S9). The MR‐Steiger directionality test suggested that there is no inverse directionality in this study (all P < 0.05). The leave‐one‐out analyses indicated that the causal estimates of MDD and the risk of infectious diseases were not driven by any single SNP (Figure S2).

Horizontal pleiotropy is a significant challenge for MR, in that genetic variation affects outcomes through other pathways than the exposure of interest. We performed multiple sensitivity analyses to minimize heterogeneity and pleiotropy. The MR-Egger intercept test showed no horizontal pleiotropy in our analyses, suggesting that our results are reliable.

 

[jnmaciuch]

Ah, are you saying some of that filtering is implied in an MR study and was probably done here? I was assuming there'd be explicit details about their methods for that.

I believe that process is what “two-stage least squares regression” refers to in their methods, though I may be wrong about that. If they didn’t do that step, it just simply wouldn’t be an MR study. I couldn’t imagine how that would get published, though I suppose it’s always possible.

Added: though as you note in the comparison to the depression MR study, it seems like they didn’t do additional tests beyond that to double check assumptions. This thread’s paper seems to be meeting bare minimum for that assumption.

 

[jnmaciuch]

Okay yes, two-stage least squares (2SLS) regression does work by estimating the causal effect as a ratio between the effect of the allele on the exposure and the effect on the outcome.

So in effect, you're starting out by looking all the alleles which are associated with CMV from the GWAS, and then regressing out the degree to which those same alleles are predictive of back pain on their own. What you're left with is the degree to which the allele is predictive of back pain only when there was CMV infection. [Added:] If that degree is 0, then the allele is effectively weighted out of the prediction (though not explicitly filtered out).

It looks like linkage disequilibrium and horizontal pleiotrophy are the two biggest possible confounders here [added:] even after you do 2SLS--they tested for LD, but didn't seem to do anything additional for the latter. I'm definitely not enough of an expert to determine if there was justification to skip that in this case.

Either way, it seems like examples of more robust analyses typically include the MR-Egger intercept test for horizontal pleiotrophy, as @forestglip brings up.

Here's some additional sources I looked at to understand the 2SLS regression method better:
Mendelian Randomization as an Approach to Assess Causality Using Observational Data
Basic Concepts of a Mendelian Randomization Approach