Author: Henry Rzepa

In the previous post, I explored the so-called “impossible” molecule methanetriol. It is regarded as such because the equilbrium resulting in loss of water is very facile, being exoenergic by ~14 kcal/mol in free energy. Here I explore whether changing the substituent R could result in suppressing the loss of water and stabilising the triol.

I started (as I usually do) with a search for crystal structures, in this case containing the motif shown below (trisubstituted carbon, disubstituted oxygen and  R = H or C and any type of connecting bond), which is the species resulting from loss of R^– to form a trihydroxycarbenium cation.

This produces six hits, of which  HIWQEJ[cite]brjshd[/cite] (DOI: 10.5517/cc3k560) and UYOYUD[cite]10.5517/ccvrghj[/cite] (DOI: 10.5517/ccvrghj) are both salts of trihydroxycarbenium cation (or protonated carbonic acid) itself – the counter ion being eg AsF₆ or an iron system. So R needs to be a stable anion and two obvious groups are triflate (trifluoromethylsulfonate) or bis(trifluoromethanesulfonyl)azanide.

The triflate (R=CF₃SO₂-O) shown below has an unusually long predicted C-O bond (1.620Å), which suggests the system is already partially ionised as shown in the top diagram. An ωB97X-D calculation [cite]10.14469/hpc/14280[/cite], DOI: 10.14469/hpc/14280) reveals the species shown below is +6.6 kcal/mol higher in free energy than the one corresponding to loss of water.

Bis-triflamide (bis(trifluoromethanesulfonyl)azanide) goes further, helped no doubt by the formation of a second strong hydrogen bond between the two ions. It is now -11.8 kcal/mol lower in free energy compared to the species resulting from loss of water.

So that is my candidate for a “possible” impossible molecule. Any takers for its synthesis?

Postscript: The next higher homologue, tris(trifluoromethanesulfonyl)methanide anion + trihydroxycarbenium cation is similar to the bis-triflamide in being -12.1 kcal/mol lower than the species resulting from loss of water.

What constitutes an “impossible molecule”? Well, here are two, the first being the topic of a recent article[cite]10.1021/jacs.4c02637[/cite]. The second is a favourite of organic chemistry tutors, to see if their students recognise it as an unusual (= impossible) form of a much better known molecule.

Perhaps we could define impossible molecules into two slightly different classes.

1. The first class is a molecule which is entirely normal in terms of its structure and bonding, but just happens to be thermodynamically less stable than an isomeric form. If all mechanistic possibilities for converting it to the more stable form are eliminated, then there is no reason it should not be detected, even though it is “impossible”. By the way, quite a number of impossible molecules have been prepared using sterics  (t-butyl groups and the like, a strategy first used perhaps 40 or so years ago) to prevent the molecule from either reacting with itself or with other molecules.
2. The second class is a molecule where the bonding or its structure are so deviant from accepted theories of the structures of molecules that its energy is either so high that either it simply cannot be prepared in the first place, or that nothing can be done to prevent its rearrangement to a much more stable form.

The first of the examples below falls clearly into the first category; methane triol. As reported[cite]10.1021/jacs.4c02637[/cite], this impossible molecule has now been detected both at low temperatures and in the gas phase at low pressure using time-of-flight mass spectrometry and other elegant experiments. The key is to ensure either a very low temperature or the absence of any acid catalyst to decompose it to methanol and formic acid.

As is my usual practice in discussing any interesting molecule, I first tend to conduct a search of the CSD (Cambridge structure database) – in this case it has to be said with little hope of finding any examples. I was therefore very surprised to find one example reported, COLRUT.[cite]10.1021/acs.orglett.9b02161[/cite] The crystal structure of COLRUT can be viewed here.[cite]10.5517/ccdc.csd.cc22yztvv[/cite] (DOI: 10.5517/ccdc.csd.cc22yztvv).  Clearly, given the discussion at the top, alarm bells should be ringing about this result. When any such alarms sound, it is my second practice to turn to calculations for verification. In this case to FAIR Data calculations[cite]10.14469/hpc/14236[/cite]  (DOI: 10.14469/hpc/14236).

The article[cite]10.1021/jacs.4c02637[/cite] also reports such calculations, but its good to have independent verification (of some of them), so I list the essential conclusions from my own calculations here.

1. At the CCSD(T)/Def2-TZVPP level, methane triol is ΔG₂₉₈ 14.49 kcal/mol higher in free energy than formic acid and water. This is not really an impossibly higher energy, and the molecule is “impossible” only because there is a very facile reaction for it to undergo (acid catalysed disproportionation for example).
2. At the much faster ωB97X-D/Def2-TZVPP level, the value is 14.48 kcal/mol, which is agrees well enough with the previous to use this method to explore further.
3. If the C-H is replaced by C-CF₃ (again a good tutorial question for how to stabilize the diol form of eg acetone), the energy of the triol is reduced to +9.4 kcal/mol. Still positive, but much smaller than the original.
4. If the C-H is replaced by C-(CF₃)₃ it is still unstable by 13.6 kcal/mol. Not much chance of using substituents to create a “possible” triol then.
5. Next, the transition state for unimolecular decomposition to water and formic acid. An IRC for this is shown below and the free energy of activation is +36.6 kcal/mol. This proceeds via a very non-linear hydrogen transfer, a geometry known to be unfavourable and indeed an energy too high for this rearrangement to occur (in a mass spectrometer? What is the temperature of molecules under these conditions?). Note how a nice hydrogen-bonded form of the products forms at the end.
   
   
   I could not resist showing the dipole moment response along the IRC. Lovely!
6. What about an intermolecular rearrangement, which would occur at either higher pressures or perhaps higher temperatures? Now, ΔG^‡ = 26.7 kcal/mol, a more viable thermal reaction.The lower barrier is because the 6-ring transition state now allows a less bent hydrogen transfer.
   
   
7. This is the reaction of a trimer, ΔG^‡ = 24.2 kcal/mol. The 8-ring transition state now allows almost linear hydrogen transfers. Note that all three transferring hydrogens move more or less in synchrony.
   
8. The tetramer: ΔG^‡ = 24.1 kcal/mol, now via a 10-ring transition state. If you look carefully at the animation, you can now see that the hydrogen transfers have become very non-synchronous (and the transition state more ionic), although they remain almost linear.
   
9. But wait, there is another isomer of the tetramer reaction, instead proceeding via an 8-ring TS, with the fourth triol molecule bonding to the transition state via four hydrogen bonds. This is very much like a stabilised protein transition state and overcomes the extra entropy of adding that fourth molecule and then some; ΔG^‡ = 18.9 kcal/mol. So at high concentrations the disproportionation of methane triol is predicted to become a facile reaction and now can only be prevented at low temperatures!<

An NCI (non-covalent-interaction) analysis of the hydrogen bonds in this TS structure is shown below. The blue regions are hydrogen bonds. The ones labelled 1-4 are the four such interactions resulting from addition of a fourth molecule to the hydrogen transfer structure of the trimer. Click for a 3D rotatable model.

So I hope this extended analysis of what makes an “impossible molecule” actually possible adds another dimension to the original report.[cite]10.1021/jacs.4c02637[/cite] As for that crystal structure, I will report to CCDC that it may in fact be an artefact and that they should take another look at the crystal structure data and correct it if needed. It is also interesting to explore the properties of cyclic hydrogen transfer reactions. The conclusion here is that an 8-ring transfer may be optimum, especially if it can be stabilized with four or more hydrogen bonds!

In an earlier post, I discussed[cite]10.59350/dfkt5-k2b20[/cite] a phenomenon known as the “anomeric effect” exhibited by tetrahedral carbon compounds with four C-O bonds. Each oxygen itself bears two bonds and has two lone pairs, and either of these can align with one of three other C-O bonds to generate an anomeric effect. Here I change the central carbon to a boron to explore what happens, as indeed I promised earlier.

One can identify candidates for such molecules by a constrained search of the CSD or the Cambridge structural database, as shown below.

The four B-O distances for each compound matching the query are now subjected to further analysis, the greatest and least values are identified and the difference between them calculated.

The results are shown in the diagram below. Three outliers are identified for close inspection.

Each of the three candidates is also subjected to a Gaussian calculation (MD15L/Def2-TZVPP)[cite]10.14469/hpc/14092[/cite] (See DOI: 10.14469/hpc/14092)

1. QIXREW[cite]10.1039/C4DT00080C[/cite]. This molecule is overall neutral and for which Δr_B-O = 0.193Å (MN15L/Def2-TZVPP Δr_{B-O =} 0.175Å). The Wiberg bond indices of longest and shortest B-O bonds are 0.486 and 0.698, Δ = 0.212Å.This is significantly larger than the best example of the C-O series, for which the largest Δr_C-O = 0.074Å and 0.137 for the Wiberg index.
2. XOVZOY[cite]10.1021/ja8071435[/cite] is a tri-anion with intercalated Ir³⁺ counterion. Δr_B-O = 0.347Å. A calculation on the isolated tri-anion (with a continuum water field to help emulate the crystal environment) results in the maximum B-O bond length difference of only 0.004Å, which is dramatically different from the crystal structure. This may be an example where the counter-ion is especially important for modelling structure, or it may be simply an anomalous refinement of the crystal structure.
3. KBDCTB, Δr_B-O measured = 0.451Å, Calculated 0.0314Å.
   This is another structure where all may not be what it seems. This again is an anionic structure and geometry optimisation of a single molecule results in a dramatic change in the internal hydrogen bonding of the species. In the crystal structure, the carboxylic acid groups all form intermolecular hydrogen bonds. Optimized as an isolated molecule, the former are no longer possible and a big conformational change occurs to allow all four carboxylic acid groups to instead form intramolecular H-bonds. In this conformation, all four B-O bonds are essentially the same length. So this might well be an example of a large change in anomeric effects due to changes in geometry induced by hydrogen bonding.

   Intermolecular H-bonds	Intramolecular H-bonds

One lesson one always learns when comparing the lengths of bonds observed in crystal structures with those calculated using quantum mechanics is that they sometimes do not match well. These mis-matches can occur for various reasons; changes in hydrogen bonding, or the presence of unmodelled counterions or simply errors in the reported crystal structure. But we might suggest from this brief foray into B-O bonds that the anomeric effects found there may indeed be larger than those of their C-O counterparts.

In the mid to late 1990s as the Web developed, it was becoming more obvious that one area it would revolutionise was of scholarly journal publishing. Since the days of the very first scientific journals in the 1650s, the medium had been firmly rooted in paper. Even printed colour only became common (and affordable) from the 1980s. An opportunity to move away from these restrictions was provided by the Web. Early adopters of this medium in chemistry were the CLIC pilot project[cite]10.1080/13614579509516846[/cite] in 1995 and the Internet Journal of Chemistry (IJC), the latter offering “enhanced chemical publication which permits the publication of materials which cannot be published on paper and end-use customization which permits the readers to read articles prepared for their specific needs“.[cite]10.1590/S0100-40421999000200020[/cite] Publication of the latter started in January 1998, offering “authors the opportunity to enhance their articles by fully incorporating multimedia, large data sets, Java applets, color images and interactive tools.” The journal remained online for seven years, after which it was closed and the articles became inaccessible. By then many major chemistry journals had started evolving along some of the same lines, and it could be argued this journal had served its purpose of alerting both publishers and authors to these new opportunities. Here I describe how an IJC article published in 2001 was brought back to life in more or less the enhanced manner intended.[cite]10.59350/9c769-34y25[/cite]

Entitled “The Mechanism and Design of Asymmetric Co-Arctate Br⁺ (Mobius) Atom Transfers Between Alkenes. A Computational Study“, an abstract of the article is still visible via services such as e.g. Scifinder, but a more complete and open metadata description which can be provided from an assigned DOI (Digital object identifier) is not available, since back in 2001 the adoption of DOIs by journals was still in its infancy. Fortunately, the original source was still available from the authors as a combination of HTML, image files and data, the latter two being hyperlinked into the body of the article. These files are in fact all that is needed to recreate the original IJC article (if not its style), using the mechanism of a data repository[cite]10.17616/R3K64N[/cite],[cite]10.25504/FAIRsharing.LEtKjT[/cite] rather than that normally designed for a journal. The procedure adopted was as follows:

1. All the data files were uploaded to the repository as a dataset.[cite]10.14469/hpc/13929[/cite], DOI: 10.14469/hpc/13929.
2. The metadata record generated and registered for these depositions (https://data.datacite.org/application/vnd.datacite.datacite+xml/10.14469/hpc/13929) has Access (the A of FAIR) identifiers in the form of e.g.
   1. <relatedIdentifier relatedIdentifierType="URL" relationType="HasPart">https://data.hpc.imperial.ac.uk/resolve/?doi=13929&file=1</relatedIdentifier>
   2. Descriptive metadata providing further properties if needed, such as file names and media types and file sizes can be obtained via
      • <relatedIdentifier relatedIdentifierType="URL" relationType="HasMetadata">https://data.hpc.imperial.ac.uk/resolve/?ore=13929</relatedIdentifier>
   3. These access identifiers replaced the hyperlinks in the original article HTML
      1. Originally: <a href="supplemental/3-ts-rh3.pdb">-1.5</a>
      2. Becomes: <a href="https://data.hpc.imperial.ac.uk/resolve/?doi=13929&file=54">-1.5</a>
      3. It is worth noting that there are basically two methods of accessing a file. The first relies on its relative path in a hierarchical file system. Hard-coding such a location into a URL means it may not be persistent – the hyperlink is vulnerable to “link rot” when the file system is reorganised and the path to the file changes. The second method relies on a database query, which should be rather more persistent, since the database should always incorporate any reorganisation of the internal systems. A third option (not used here) is to assign a persistent identifier to every file, and to ensure that a properly persistent direct access mechanism is described in metadata for that file.
      4. The root document for the article, given the reserved filename index.html was edited to reflect the changes in the hyperlinks.
3. The article document index.html was now itself uploaded to the repository. In a conventional data repository, such a file invokes no specific actions, but in the repository used for this purpose it does have the reserved meaning of invoking in effect a preview or “LiveView” using the syntax
   • <iframe name="liveview" src="https://data.hpc.imperial.ac.uk/resolve/?doi=13929&file=90"
     width="100%" height="600" id="liveview"></iframe>
4. The article now functions much in the same way it would have done on IJC, albeit in one interesting way. The regular style adopted in journals is to place the ESI or electronic supporting information files into a separate enclave, linked via the article landing page by parochial mechanisms. In this instance the article and its data files are visible on the same page – it is a data repository after all – thus elevating the data to the same status as the article. Such elevation is often referred to as making “Data a first class citizen of the publication processes“.
5. The opportunity now arose to incorporate an interactive tool based on the use of the JSmol molecule viewer.
   • By adding an additional header to the HTML document containing a Javascript invocation of JSmol, selected data could be brought to life by creating a molecular model in a separate window.
   • This is invoked by a variation on the hyperlink shown above in section 3.2 by
     <a href="javascript:show_jmol();javascript:handle_jmol('10.14469/hpc/13933',%20';frame 1;font label 16;zoom 5;moveto 4 90 4 80 65 120;spin 3;set echo bottom left;font echo 20 serif bolditalic;color echo green;echo TS for 3 (C2 symmetry);')">Load 3D Model</a>
   • Additional tools are now provided, from activating a (molecular) vibration, calculating a chirality (if applicable) or others invoked from a pull-down menu.
   • In this example, the data is again accessed directly from a data repository, albeit by a different mechanism from that shown in 3.2 and here based only on the DOI of the data and its media type (in this case chemical/x-mdl-molfile).

It was not the intention here to illustrate how a Journal infrastructure might work – merely to rescue an article published 23 years ago (a long time in the Internet era) from a journal that is no longer disseminating articles. In the process the article has acquired its own DOI (albeit as data and not journal article), something not available from the original journal and some level of interactivity of the type originally envisaged. The (manual) process took something around 2-3 hours to achieve, and would certainly need automating if it were to be used more than once. I take encouragement however that after so many years, it was still possible with relatively little effort to achieve this curation.

The recent release of the DataCite Data Citation corpus, which has the stated aim of providing “a trusted central aggregate of all data citations to further our understanding of data usage and advance meaningful data metrics” made me want to investigate what the current state of citing data in the area of chemistry might be. Chemistry is known to be a “data rich” science (as most of the physical sciences are) and  here on this very blog I try to cite whenever possible the source(s) of the data that  I often use when discussing a topic. Such citations are not necessarily the same as citing a journal source via e.g. its DOI, although of course one is very likely to find data associated with most articles nowadays, albeit almost entirely via any associated supporting information document. However the latter is often presented in a relatively unstructured (PDF) form, which does not adhere to what are called the “FAIR” guidelines of being findable, accessible, interoperable and reusable. Directly citing data is a way of improving its FAIR-characteristics. So what insights does the Data citation corpus reveal?

1. This overview shows that by far the most common mechanism for citing data is via its Accession Number, used predominantly by Life Sciences (an example of this latter is linked here[cite]10.1038/s41597-022-01707-6[/cite]), with the DOI (digital object identifier) being less common.
   
2. Tunnelling down to citation counts in chemical sciences by publisher, an odd picture emerges with just a handful of citations.
   
3. The more general physical sciences does not fare much better:
4. Lets try a different approach, filtering by repository. Thus here are the statistics for the Cambridge crystallographic data centre, which was citing data in large amounts a few years back, but which appears to have dropped off in the last few years. Given that the entries there continue to go up almost exponentially, we begin to suspect that the data citations there are not being properly recognised as such by the citation corpus.
   
5. Lets try another repository, Zenodo, which again is dropping but where the totals are about 500 a year for the most recent.
6. OK, one more go, the RSC chemistry publisher.
   

I am not sure what to make of this; areas where you would expect very high levels of data citation in chemical sciences do not appear to exist – I think for some reason, the DataCite citation corpus is not yet capturing them.[cite]10.59350/t80g1-xys37[/cite] But when things do start operating as perhaps expected, I think we will have a very valuable resource, which should firmly put data (whether FAIR or not) on the map.

In the late 1980s, as I recollected here[cite]10.59350/g4j62-4xk50[/cite] the equipment needed for real time molecular visualisation as it became known as was still expensive, requiring custom systems such as Evans and Sutherland PS390 workstations. One major breakthrough in making such techniques generally available on less specialised equipment was achieved by Roger Sayle[cite]10.1016/s0968-0004(00)89080-5[/cite], then working at Imperial College around 1990 and using a Silicon Graphics workstation. He greatly optimised up the rendering algorithms by creating a program called RasMol (after his initials), which meant such visualisations could very rapidly also be achieved even on a personal computer. Moving from vector display technology (the PS390) to Raster/bitmap graphics had allowed spacefilling representations of molecules containing 100s if not 1000s of atoms – and in turn enabled the new World-Wide Web to exploit the technique.[cite]10.1039/C39940001907[/cite]

Whilst Rasmol is very much still around, it also provided an inspiration for successor programs such as Jmol (based on Java) and JSMol (based on the Javascript language built into all modern web browsers). There are now many articles in the literature describing this program. In 2008 the very first post on this blog described how run it in a WordPress instance[cite]10.59350/pq7ds-gqr71[/cite].

Now a new milestone in molecular visualisation has been reached – the ability to display 3 million atoms! Bob Hanson has just released Jmol/JSmol 16.1.51 which supports the BinaryCIF file format. An example of the power of both program and this new format is illustrated with the protein 8glv[cite]10.2210/pdb8GLV/pdb[/cite] which contains 3 million atoms (the bcif file itself is only 47.4 Mb).

The Jmol/JSmol script to load it is:

t = now();    
     set autobond false;
     load =8glv.bcif filter "*.CA";
     spacefill on;
     color chain;
     print now(t);

and the actual rendering takes just 10-20 seconds. You can see from the screenshots below^‡ that when it is zoomed in, it really does show individual atoms! Who knows what the practical atom limit is, but it is almost certainly more than three million! And it may even be possible on a mobile phone!

^‡OK, you are asking why I have not loaded 8glv into this page? Well, I need to update JSmol on this site first and have encountered an issue that needs fixing.

Postcript. We are getting there. Below is a screen capture of this protein using an iPhone 15 and any of Safari, FireFox, Chrome or Edge browsers via this link (ca 12 seconds on iPhone, ~30 seconds on a 2019 iMac, ~12 seconds on an M1 Mac Studio). It even spins reasonably smoothly!