Generics (UK) Limited & Ors v AstraZeneca AB

The judge approached the issue of plausibility by asking himself the three questions proposed by Birss LJ in Akebia Therapeutics Inc v FibroGen Inc [2021] EWCA Civ 1279, [2022] RPC 7 at [53] for addressing claims to classes of products:

“First one must identify what it is which falls within the scope of the claimed class. Second one must determine what it means to say that the invention works. In other words what is it for? Once you know those two things, the third step can be taken: to answer the question whether it is possible to make a reasonable prediction the invention will work with substantially everything falling within the scope of the claim.”

It was common ground that the answer to the first question is that claim 2 is limited to the single compound dapagliflozin, while claim 15 is limited to the use of that compound in the manufacture of a medicament for treating or delaying the progression or onset of diabetes (“treating diabetes” for short).

It was also common ground that the answer to the second question in the case of claim 15 is that “working” means treating diabetes. That is the relevant technical effect. As for claim 2, AstraZeneca did not advance any case that the identification of dapagliflozin per se was a technical contribution (i.e. without identifying any useful properties of dapagliflozin). The judge concluded at [232] that:

“… what it means to ‘work’ for the purpose of the second Fibrogen step in respect of claim 15 is treating etc. diabetes, in the sense of having sufficient efficacy to treat diabetes, and in respect of claim 2 is either (a) the same as in the case of claim 15 or (b) reducing blood or plasma glucose in vivo with sufficient efficacy to allow it to be used as an experimental tool.”

On the appeal AstraZeneca confined its case to finding (a), which made it unnecessary for the Claimants to pursue a respondents’ notice challenging finding (b).

The judge considered the third question at [233]-[262]. He began by noting that Prof Thorens and Prof Bailey were in agreement that the mechanism proposed in the Patent, i.e. that SGLT2 inhibition would lead to a reduction of blood/plasma glucose in vivo and hence an effect on the diabetes disease state was a plausible one, given the support provided by the phlorizin and Tanabe Seiyaku studies referred to in [0004] and [0007]. The question was whether the Patent disclosed enough to make it plausible that dapagliflozin has activity as an SGLT2 inhibitor which is sufficient to have a useful effect on the diabetes disease state.

The issue between the parties was whether the statements in the Patent referring to dapagliflozin as an SGLT2 inhibitor were a bare assertion, as the Claimants contended, or verbal statements of an experimental result, albeit without numerical data, as AstraZeneca contended. The crux of this dispute was the correct interpretation of [114] and [115].

So far as [114] was concerned, the judge interpreted this as a statement that the assay described in [115] could be performed to determine SGLT2 activity.

As for [115], it was common ground that, given that this was expressed in the past tense, it was to be taken as a description of an assay that had been carried out. The judge noted, however, that what was used in the assay was “inhibitor”. There was nothing to tell the reader what the inhibitor was. The judge went on:

“239.

Further, I accept the Claimants’ submission, supported by their cross-examination of Prof. Bailey, that the skilled person reading the Patent would understand that the assay could have been performed by the patentee on other compounds, either in the validation of the assay, in which case phlorizin might have been used as the inhibitor, or in assessing the SGLT2 inhibitory activity of other compounds. The fact that paragraph [0115] tells the reader that the assay has been performed on an inhibitor does not tell the reader that it has been performed on dapagliflozin.

240.

The reader might guess that had been done, but that is not the same as a disclosure that it has been. While Prof. Bailey and Prof. Potter said that the reader would understand that the assay had been performed on dapagliflozin, that was based on their assumption that a pharmaceutical company would not apply for a patent (let alone be granted one) if it did not have data to support the assertion that dapagliflozin was an SGLT2 inhibitor. That is a bootstraps argument, and in my judgment the assumption is not a permissible one to make …. Further, both Dr Edwards and Prof. Potter said that the fact that 20g dapagliflozin had been made did not make any difference to the assessment of whether it had been assayed.

241.

Overall, I reject AZ’s characterisation of the Patent's description of dapagliflozin as an SGLT2 inhibitor as a verbal statement of an experimental result. On the contrary, it is an assertion unsupported by any experimental results.

242.

As Warner-Lambert makes clear, a priori reasoning can provide an alternative basis for plausibility of an effect. However, the Patent contains no reasoning, based on structural similarities or otherwise, to support the suggestion that dapagliflozin is an SGLT2 inhibitor.”

The judge explained that AstraZeneca relied on structural similarities between dapagliflozin and phlorizin as giving the skilled team “additional reassurance” that dapagliflozin would be an SGLT2 inhibitor. Having identified five differences between the structures of dapagliflozin and phlorizin, the judge went on:

“245.

It was quite clear from the evidence of Dr Edwards and Prof. Potter that, given the CGK that small structural differences can make large differences to activity, the number and nature of the differences between dapagliflozin and phlorizin were such that no prediction could be made, based on the activity of phlorizin, about the activity of dapagliflozin as an SGLT2 inhibitor. The most that could be said was that the removal of the O-glucoside bond might reduce the risk of hydrolysis of dapagliflozin compared to phlorizin, and that the glucose moiety in dapagliflozin might interact with SGLT2. However, I accept Dr Edwards’ evidence that the presence of the glucose moiety, together with the two phenyl rings, does not provide the skilled team with the ability to make a reasonable prediction that dapagliflozin would be an SGLT2 inhibitor.

246.

Indeed, AZ accepted that the similarities (such as they are) with phlorizin would not on their own be sufficient to make plausible the assertion that dapagliflozin was an SGLT2 inhibitor. It relied on them only to ‘reinforce the confidence’ that the skilled team would be given by ‘the positive assay result’. However, in my judgment there is no ‘positive assay result’ disclosed in the Patent for the reasons explained above.”

Next, the judge explained that there was a debate between the parties as to whether the skilled team when presented with the Patent would obtain and study any of the documents referred to in it, and if so which. The judge directed himself that this was “a context- and fact-dependent question, and thus it depends firstly upon the wording of the specification and secondly on the evidence” (Akebia Therapeutics Inc v FibroGen Inc [2020] EWHC 866 (Pat), [2020] RPC 15 at [218]).

The judge went on:

“248.

Prof. Potter’s view was that the skilled medicinal chemist reading the Patent would see the reference to WO 128 at paragraph [0011] and look it up. I accept that - in my judgment a skilled medicinal chemist presented with a statement that dapagliflozin is an SGLT2 inhibitor but no experimental data to support it, and a statement in paragraph [0011] that compounds of a Markush formula which covers dapagliflozin have been reported in WO 128 to be SGLT2 inhibitors, would want to obtain WO 128 to see whether it contained SAR information that could be used to assess the claim of SGLT2 inhibition for dapagliflozin. Of course, if the skilled medicinal chemist were to obtain WO 128, they would find that it too contains no experimental data, and therefore no SAR information which might make it plausible that dapagliflozin was an SGLT2 inhibitor, let alone indicate what its EC₅₀ might be.

249.

Prof. Thorens and Prof. Potter said that the skilled team presented with the statements in paragraphs [0004] and [0007] of the Patent about the Tanabe Seiyaku 6 month rat study (or their equivalents in WO 128) would want to find out more about that work. However, no attempt was made to recreate the kind of search that would have been done at the priority date to locate that work. The skilled team which had WO 128 would, however, see that it contained a number of references to prior work on compounds which are said to be SGLT2 inhibitors. Those references include the papers by Tsujihara et al., Hongu et al. and Oku et al. …; in my judgment the skilled team would regard those as potential sources for the Tanabe Seiyaku work referred to in paragraphs [0004] and [0007]. Further, Prof. Potter said that the skilled medicinal chemist would want to look at materials cited in WO 128, including those papers, and would be interested in any SAR information they contained. Therefore, in my judgment the skilled team which had WO 128 (which for the reasons explained in the previous paragraph, the skilled team reading the Patent would have done) would also look at the four Tanabe Seiyaku papers which it cites.

250.

Even a brief consideration of those papers would reveal that the SAR work conducted by Tanabe Seiyaku indicated strict structural requirements for activity of phlorizin analogues. The position can be summarised by setting out the abstract of the first Hongu et al. paper (which follows on from the Tsujihara et al. paper which had shown a benefit of removing the 4'-hydroxy group from the first phenyl ring, referred to as the B ring by Tanabe Seiyaku):

‘A novel series of 4'-dehydroxyphlorizin derivatives was synthesised and the effects of these compounds on urinary glucose excretion were evaluated in rats. There was a strict structural requirement for activity. Introduction of a small substituent or a flat ring at the 3- and/or the 4-position on the A ring was permissible, but any change at the bridge part between the A and B rings or in the sugar moiety resulted in complete lack of activity. The 6'-OH group on the B ring was also necessary, and even small structural modifications of the 6'-OH group reduced the activity considerably. Among the compounds synthesised, the 5-benzofuryl derivative 25 [T-1095a] was the most potent and was selected as a new lead for further structure-activity relationship investigations.’

251.

This emphasises the importance, in phlorizin analogues, of retaining the structure of the linker between the two phenyl rings, and the hydroxy group on the first phenyl ring that has the ability to hydrogen bond to the carbonyl in the linker. Both those features are absent in dapagliflozin, and so the skilled team considering the Tanabe Seiyaku papers would find nothing to suggest that dapagliflozin would be an SGLT2 inhibitor. Indeed, in my judgment the message that they convey, namely that structural features of phlorizin analogues (which are absent from dapagliflozin) are important for activity, would give the skilled team reason to doubt whether dapagliflozin would be an SGLT2 inhibitor.

252.

As neither WO 128, nor the Tanabe Seiyaku papers which it cites, contain any information to support the suggestion that dapagliflozin is an SGLT2 inhibitor, it is not necessary to consider whether it would be permissible to rely on information in these cross-referenced documents to establish or support a case of plausibility applying the approach in Warner-Lambert.”

Finally, the judge considered the question of potency:

“253.

AZ repeatedly asserted that the teaching of the Patent was that ‘positive results’ had been obtained in the assay of dapagliflozin. However, even if one were to accept (which I do not) that the teaching of the Patent was that dapagliflozin had been assayed, that begs the question as to what the ‘positive results’ were. Even if the skilled person were to understand the Patent as disclosing that dapagliflozin had been assayed and determined to be an SGLT2 inhibitor, it does not disclose the EC₅₀ that had been obtained (or even how it compared with that of phlorizin).

254.

Nor, on the evidence, did the description of the assay mean that any positive result must have been a EC50 in a range which would confer utility. Dr Edwards said, and Prof. Bailey accepted, that the potency could have been in the millimolar range. AZ submitted that the assay could have been one with a 10 μM EC₅₀ cut-off, and that a compound which passed such an assay would have been a ‘hit’. I do not think there was anything in the evidence to suggest that the skilled team would have understood the assay to have such a cut-off, but in any event, on the evidence a ‘hit’ in itself is not a compound with practical utility ….

255.

AZ submitted that the statements in the Patent that dapagliflozin was an SGLT2 inhibitor useful for the treatment of diabetes must be read as teaching that the EC₅₀ for dapagliflozin was such that it would be useful to treat diabetes. I agree with the Claimants’ characterisation of this as another bootstraps argument. On the contrary, the Patent makes bare assertions that dapagliflozin is an SGLT2 inhibitor and useful to treat diabetes.”

Having quoted two passages from the cross-examination of Prof Bailey in [256], the judge went on:

“257.

Overall, the evidence was that not every compound which produced a measurable EC₅₀ in an in vitro SGLT2 inhibition assay would have been regarded as plausibly having a useful effect on … the diabetes disease state. Indeed, while my conclusion does not depend on this, the skilled team which had the Tanabe Seiyaku papers cited in WO 128 would, as Prof. Potter accepted, note that Tsujihara et al. presents results showing that some compounds which had SGLT2 activity in vitro were indistinguishable from the control in vivo and so were categorised as inactive.

258.

The skilled team which had the Tanabe Seiyaku papers cited in WO 128 would also note that Oku et al. provides the IC₅₀s for phlorizin (160 nM against both SGLT1 and SGLT2) and for [a compound designated] T-1095a (20 nM against SGLT1 and 5 nM against SGLT2). It also presents the results of a study of T-1095a in diabetic rats, though it does not appear to be the study referred to in paragraphs [0004] and [0007] of the Patent (the source of that study was not clear from the materials in the case). There is nothing in the Tanabe Seiyaku papers to displace the skilled team’s CGK about potency requirements for an SGLT2 inhibitor which could be useful to treat diabetes. They provide the IC₅₀ for phlorizin as a benchmark for someone seeking an experimental tool to lower blood/plasma glucose. However, as I have said, the Patent provides no information on the EC₅₀/IC₅₀ of dapagliflozin.

259.

The absence of information about the EC₅₀ of dapagliflozin is significant. The skilled team cannot reasonably predict any useful effect on … the diabetes disease state from merely being told that dapagliflozin is an SGLT2 inhibitor.

260.

For the reasons I have given above, in my judgment the Patent does not disclose enough to make it plausible that dapagliflozin … will treat etc. diabetes.”

The judge explained at [266]-[268] that AstraZeneca’s case at trial was that it was not necessary for dapagliflozin to have greater efficacy than (or be in any other way superior to) any of the other compounds encompassed by the disclosure of WO 128, including the compounds of the Examples. AstraZeneca did not advance any case that dapagliflozin was, so far as utility for the treatment of diabetes was concerned, any different in its properties to any of the other compounds of WO 128 (and in particular to the compound of Example 12).

In those circumstances the judge held as follows:

“269.

…. in my judgment that means that the Patent does not make a technical contribution over WO 128. There is nothing in the Patent nor in the evidence at trial to indicate that dapagliflozin is anything other than an arbitrary selection from the genus of compounds disclosed by WO 128.

270.

The argument advanced by AZ was that the Patent made it plausible that dapagliflozin was an SGLT2 inhibitor which … was useful to treat etc. diabetes, whereas that was not made plausible by WO 128. That, AZ submitted, taught the skilled person something new and so was a technical contribution.

271.

In my judgment there is some sleight of hand going on here. A technical contribution needs to be both present in fact and made plausible by the specification ….. I do not see how making something plausible can in itself be a technical contribution. Indeed AZ did not plead a technical contribution consisting of making something plausible - it pleaded (correctly, in my judgment) actual properties of dapagliflozin as being technical contributions.

272.

Even if this argument were a permissible one, in my judgment it does not work in this case. The parties were agreed that, in order for the arbitrary selection case to add anything to the plausibility case, it is necessary to assume that the Patent does make it plausible that dapagliflozin is an SGLT2 inhibitor which … is useful to treat etc. diabetes. That requires me to approach matters on the basis that the Patent would be understood as disclosing that dapagliflozin had been assayed and had generated an EC₅₀ which meant that it would plausibly … be useful to treat etc. diabetes.

273.

If that is so, then in my judgment WO 128 must also be understood as disclosing that compounds which it discloses had been assayed and had generated EC₅₀s which meant that they would plausibly … be useful to treat etc. diabetes. Prof. Potter (whose view was that the skilled medicinal chemist reading the Patent would assume that dapagliflozin had been tested) said this:

‘While the Medicinal Chemist would not be able to identify which of the compounds encompassed by WO 128 were tested in the assay, they would not doubt that the assay had in fact been used in relation to some of them. While no biological data are provided, it would be a reasonable assumption that at least the Examples - for each of which characterisation data are provided, indicating that they were synthesised - would have been tested for activity using this assay.’

274.

Therefore, on this approach the skilled team would understand that the compound of Example 12 of WO 128 (amongst others) had been tested and produced an EC₅₀ which meant that it would plausibly … be useful to treat etc. diabetes.”

The judge went on to refer to Prof Potter’s evidence that the skilled medicinal chemist would regard the statement in WO 128 that the compounds of Formula IB are SGLT2 inhibitors and useful in the treatment etc. of diabetes as being a prediction based on assay results of the compounds of the Examples, and continued:

“277.

There was no suggestion that any particular compounds covered by formula IB would be regarded as likely exceptions to the prediction - in particular there was no suggestion that the skilled medicinal chemist would regard a compound with R¹ = Cl and/or R⁴ = OEt as being a likely exception. On the contrary, Prof. Potter’s evidence was that the skilled medicinal chemist would have no expectation that a compound with an alternative alkoxy group would have better, the same or worse activity data than one with a methoxy group (as in Example 12).

278.

Therefore, on the hypothesis on which I am considering matters, WO 128 contains what the skilled team would regard as a plausible prediction, based on assay data for the compounds of the Examples, that compounds of formula IB (including dapagliflozin with its close relationship to Example 12) will be SGLT2 inhibitors and that they will … be useful to treat etc. diabetes.

279.

All that the Patent does (on that hypothesis) is to verify that WO 128’s prediction of SGLT2 inhibitory activity and utility is correct in the case of dapagliflozin. It does not show that dapagliflozin has properties that are in any way different from those predicted by WO 128. On that hypothesis, while the Patent may make it more plausible that dapagliflozin … be useful to treat etc. diabetes, it does not make that plausible for the first time. So I reject AZ’s argument which I have recorded in paragraph 270 above.”

AstraZeneca’s grounds of appeal fall into three groups. Grounds 1-3 concern the judge’s interpretation of the Patent. Grounds 4-6 concern the law as to plausibility. Grounds 7 and 8 concern arbitrary selection.

Ground 1 is that the judge wrongly assessed the disclosure of the Patent at [238]-[241] when he found that it did not comprise a verbal statement of an experimental result. AstraZeneca relies on the principle, which is not in dispute, that the skilled person or team reads a patent with a “mind willing to understand, not a mind desirous of misunderstanding” (Generics (UK) Ltd v Yeda Research and Development Co Ltd [2012] EWHC 1848 (Pat) at [192]). Given that the Patent is all about dapagliflozin, that it contains repeated references to “the SGLT2 inhibitor of formula I” and that the purpose of the assay described in [0115] is to measure “SGLT2 activity of the compounds of the invention” as stated in [0114], AstraZeneca contends that the skilled team would understand that the “inhibitor” referred to in [0115] must be dapagliflozin.

The Claimants point out that, as Lord Diplock emphasised in Catnic Components Ltd v Hill & Smith Ltd [1982] RPC 183 at 242, “a patent specification is a unilateral statement by the patentee, in words of his own choosing”. Even so, if the issue were one purely of language, AstraZeneca’s argument might have some merit. A patent, however, is a technical document addressed to technical readers. It is to be read through the eyes of the skilled person or team having the common general knowledge. Read in that way, I have no hesitation in agreeing with the judge.

What would immediately strike the skilled team about [0115] is that it does not identify the “inhibitor”. It is purely a description of a method. If one is looking for linguistic clues, it is noticeable that it does not even refer to “the inhibitor”, which might be taken to be a reference to a specific, even if unidentified, inhibitor. It simply uses the word “inhibitor” as a generic term for the substance being tested. If [0115] was intended to record a test that had been performed on dapagliflozin, then it would surely have said so. Nothing would have been easier, and there is no good reason why that statement should not have been made, if that were the case.

Furthermore, the skilled team would note that [0115] does not itself contain even a verbal statement of a result. It does not say, for example, “the inhibitor was found to have SGLT2 activity”, let alone “significant [or suitable or some other such term] SGLT2 activity”. Again, nothing would have been easier, and there is no good reason why that statement should not have been made, if that were the case.

AstraZeneca contended in paragraph 4 of its grounds of appeal and paragraph 16 of its skeleton argument that the judge ought not to have relied upon the expert evidence to which he referred at [239] to resolve what is properly a matter of documentary interpretation for the court. I did not understand counsel for AstraZeneca to pursue this contention in oral argument, but in any event I would reject it. The judge made a finding in [239], based on the expert evidence, as to a technical consideration which would bear upon the skilled team’s reading of the Patent, namely that the assay described in [0115] could have been performed on other compounds either in validating the assay or in assessing the SGLT2 inhibitory activity of other compounds (such as one or more of those described in WO 128). It follows that the Patent must be interpreted with that consideration in mind. The judge did not make the mistake of treating the expert evidence as determinative of the issue of interpretation.

Instead, counsel for AstraZeneca focussed his argument on the proposition that the evidence did not establish that the skilled team would think that only other compounds had been tested and dapagliflozin had not. This argument makes the very mistake which AstraZeneca’s abandoned argument accused the judge of making, namely of treating the issue of interpretation as being one for expert evidence. In interpreting the Patent, what matters is that [0115] gives the skilled team no reason to think that dapagliflozin has been tested when there are alternative possibilities which are equally consistent with the language used.

Ground 2 is that the judge wrongly assessed the disclosure of the Patent at [253]-[260] when he found that the Patent did not disclose enough to make it plausible that dapagliflozin will treat diabetes. AstraZeneca contends that the judge should have found that the assay result was that dapagliflozin had been shown to be an SGLT2 inhibitor at a useful level of activity. The sole basis for this contention is the repeated references in the specification to dapagliflozin as an SGLT2 inhibitor useful for the treatment of diabetes.

This argument can only get off the ground if AstraZeneca succeeds on ground 1. Even on that assumption, however, I do not accept it. As the judge rightly said, it is a bootstraps argument. The repeated references in the specification to dapagliflozin being an SGLT2 inhibitor useful for the treatment of diabetes are pure assertion. Even if the skilled team understood [0115] to mean that dapagliflozin had been tested and found to demonstrate some unspecified degree of SGLT2 inhibition, there is no information in the Patent as to: (i) what level of SGLT2 potency was obtained, even in purely verbal terms; or (ii) more specifically, the EC₅₀ value that was determined; let alone (iii) whether the EC₅₀ value was sufficient to confer utility. The judge made findings in [254] and [257], based on the expert evidence, that the potency could have been in the millimolar range, which would not have been enough for practical utility, and that not every compound which produced a measurable EC₅₀ in an in vitro SGLT2 inhibition assay would have been regarded as plausibly having a useful effect on the diabetes disease state.

AstraZeneca’s only real answer to this was to attack the judge’s findings as being unsupported by the evidence of Prof Bailey which the judge quoted in [256]. This is not a ground of appeal for which AstraZeneca has permission. In any event, it is a hopeless argument. The evidence of Prof Bailey provided ample support for the judge’s finding and it did not stand alone, as the judge made clear. Counsel for AstraZeneca pointed out that the first passage quoted by the judge was in the context of WO 128, and only the second in the context of the Patent, but there is nothing in this point because the disclosures of the documents are the same with respect to the assay.

Ground 3 is that, had the judge correctly interpreted the patent, he should have concluded that the claimed inventions were plausible even applying the standard laid down in Warner-Lambert. This depends on AstraZeneca succeeding on both grounds 1 and 2, and therefore does not arise.

AstraZeneca accepts that, if it fails on grounds 1-3, then claim 15 is insufficiently disclosed applying the Warner-Lambert standard. On that hypothesis, AstraZeneca nevertheless contends that claim 2 is valid. In outline, AstraZeneca argues that claim 2 satisfies the requirement of inventive step because the skilled team reading the Patent at the priority date would have no legitimate doubt that dapagliflozin would be useful for the treatment of diabetes, and that the invention of claim 2 is sufficiently disclosed because, as is common ground, the Patent discloses how to make dapagliflozin.

Before considering grounds 4-6 in detail, I would point out that claim 2 is significantly broader than claim 15. It covers all uses of dapagliflozin (including for the treatment of the diseases deleted from claim 14, as well as wholly unrelated diseases, and non-pharmaceutical applications), while claim 15 is limited to the use of dapagliflozin for the treatment of diabetes. As a matter of basic principle, a broader claim to monopoly requires more justification than a narrower claim. Yet AstraZeneca contends that the patentability of claim 2 should be subject to less stringent criteria than claim 15. Moreover, this contention is advanced even though precisely the same disclosure in the Patent is relied upon as supporting both claims. Counsel for AstraZeneca was unable to provide any convincing justification for this.

I would also point out that, as the judge recorded, AstraZeneca does not contend that the identification of dapagliflozin was a patentable invention. This is because the mere identification of a new chemical compound requires no invention: see e.g. T 939/92 Agrevo/Triazoles [1996] EPOR 171 at [2.5.3], Pharmacia Corp v Merck & Co Inc [2001] EWCA Civ 1610, [2002] RPC 41 at [20] (Aldous LJ) and Idenix Pharmaceuticals Inc v Gilead Sciences Inc [2016] EWCA Civ 1089 at [116] (Kitchin LJ). AstraZeneca’s argument for patentability rests upon the proposition that dapagliflozin is useful for the treatment of diabetes. Thus AstraZeneca simultaneously argues that (i) what makes the difference between claim 2 and claim 15 is that claim 2 does not include the therapeutic effect of dapagliflozin as a functional feature of the claim whereas claim 15 does, and (ii) the invention of claim 2 is patentable because of the therapeutic effect of dapagliflozin. Again, counsel for AstraZeneca did not provide any convincing justification for this.

Ground 4 is that the Warner-Lambert test sets too high a standard for plausibility, and must be revisited in the light of G 2/21. Ground 5 is that, in the light of G 2/21, the Court should have treated claim 2, as a claim to a product per se, differently to claim 15, as a claim comprising a functional technical feature. The plausibility of the former should be assessed under inventive step, and the plausibility of the latter under sufficiency of disclosure, with different standards for plausibility applying in those cases: ab initio implausibility in the former case and ab initio plausibility in the latter case.

The short answer to ground 4 is that, as I have already noted, it is common ground that Warner-Lambert is binding on this Court with respect to the sufficiency of medical use claims. Counsel for AstraZeneca did not argue that G 2/21 provided any basis for this Court to apply a different test of plausibility in such cases, and he accepted that the Enlarged Board’s (in English terms, obiter) conclusion at [77] was consistent with that of the majority in Warner-Lambert.

Accordingly, counsel for AstraZeneca concentrated his argument on ground 5. As he accepted, Sandoz v BMS is prima facie binding authority to the opposite effect. Furthermore, he did not challenge the reasoning in Sandoz v BMS at [92] quoted in paragraph 19 above. He nevertheless submitted that this Court was free to depart from Sandoz v BMS, and should do so, on the basis of the exception to the rule that a previous decision of this Court is binding on this Court (and on lower courts) recognised in Actavis UK Ltd v Merck & Co Inc [2008] EWCA Civ 444, [2009] 1 WLR 1186 at [95]: “we are free to decide what this court is to do if it finds that its earlier interpretation of the [EPC] is clearly inconsistent with a settled interpretation given by the Boards of Appeal” (Jacob LJ giving the judgment of the Court of Appeal).

As counsel for AstraZeneca emphasised, decisions of the highest authority in this jurisdiction have recognised the desirability, in the absence of any supranational court with jurisdiction to interpret the EPC, of the UK courts following the settled jurisprudence of the Boards of Appeal: see, for example. Human Genome Sciences Inc v Eli Lilly and Co [2011] UKSC 51, [2012] RPC 6 at [87] (Lord Neuberger). It is important that a harmonised approach to the interpretation of the EPC is adopted by the courts of the Contracting States even if that means the UK courts adopting an interpretation that is contrary to the interpretation they would have adopted if left to their own devices: see, for example, John Wyeth & Brother Ltd’s Application [1985] RPC 545 at 565-567.

Counsel for AstraZeneca’s principal submission was that G 2/21 is itself sufficient basis for this Court to depart from Sandoz v BMS. Although G 2/21 was considered in Sandoz v BMS, it had been common ground between the parties in that case that it made no difference whether the issue was viewed as one of inventive step or as one of sufficiency, whereas AstraZeneca contends that it makes a significant difference. As a result, G 2/21 had been misinterpreted in Sandoz v BMS. His secondary submission was that subsequent decisions of the Boards of Appeal and of courts of other EPC Contracting States confirm that G 2/21 was misinterpreted in Sandoz v BMS and supported AstraZeneca’s reading of it.

So far as G 2/21 is concerned, counsel for AstraZeneca emphasised that the Enlarged Board had expressly differentiated between inventive step and sufficiency at [73]-[77] (see paragraph 31 above). He argued that it followed that the Enlarged Board could not have intended to apply the same criteria to inventive step as to sufficiency, and in particular sufficiency of medical use claims. I acknowledge the force of this argument, but the question remains how the criteria articulated by the Board as being applicable to the assessment of inventive step are to be understood and applied.

Counsel for AstraZeneca submitted that the first limb of the test asks whether the relevant technical effect is part of what is taught by the document. If it is not, then it cannot be relied upon as a putative inventive step. The second limb is a question of technical substance in light of the skilled addressee’s technical understanding: in essence, whether the skilled addressee would understand, in light of their common general knowledge, that the subject-matter of the claim has the technical effect relied upon.

I did not understand counsel for the Claimants to dispute counsel for AstraZeneca’s reading of the first limb. As counsel for the Claimants expressed it, it is a disclosure test. Nor did counsel for the Claimants really take issue with counsel for AstraZeneca’s formulation of the second limb as set out in the preceding paragraph (which I have taken verbatim from paragraph 43 of AstraZeneca’s skeleton argument).

Where counsel parted company was with regard to counsel for AstraZeneca’s further submission that the second limb has no requirement for a positive teaching in the specification rendering the effect plausible, and is therefore akin to the ab initio implausibility test. Counsel for the Claimants disputed this.

Apart from the Enlarged Board’s differentiation between inventive step and sufficiency, counsel for AstraZeneca’s principal argument in support of this submission was that the Enlarged Board had treated the ab initio plausibility lines of cases as being reconcilable with the ab initio implausibility line of cases. He submitted that this could only be the case if one “levelled down” to the ab initio implausibility standard. Counsel for the Claimants riposted that that would mean that the ab initio plausibility line of cases were wrongly decided, and yet the Enlarged Board had affirmed them.

The conclusion I draw from these arguments, revisiting G 2/21 after an interval of two years since BMS v Sandoz, is that, while the Enlarged Board clearly intended to get away from the debates over plausibility in the context of inventive step, the truth of the matter is that the distinction between ab initio plausibility and ab initio implausibility is inescapable. The ab initio plausibility standard requires the claimed technical effect to be rendered plausible by the technical content of the application read with the skilled person or team’s common general knowledge. The ab initio implausibility standard permits a technical effect to be claimed that is not implausible in the light of the technical content of the application read with the skilled person or team’s common general knowledge. As is demonstrated by the arguments in Warner-Lambert, Sandoz v BMS and the present case, this difference can be decisive where the application claims a technical effect without providing either experimental data or theoretical reasoning to support that claim.

I would add that, as Lord Sumption pointed out in Warner-Lambert at [40], the difference between the two standards only really matters in cases where it has subsequently been demonstrated that the technical effect is indeed achieved. In cases like Sandoz v BMS and the present one, the technical effect has subsequently been proved not merely by the in vitro data that is missing from the specification, but also by in vivo experiments and then by clinical trials. In those circumstances it is tempting for courts and tribunals to be swayed by that hindsight knowledge. Tempting as it is, this is contrary to principle for the reasons explained in paragraphs 11-13 above.

I am therefore not persuaded that G 2/21 itself justifies this Court departing from Sandoz v BMS. I turn next to consider subsequent decisions of the Boards of Appeal and of the national courts.

So far as decisions of the Boards of Appeal are concerned, AstraZeneca primarily relies upon the decision of the referring Board of Appeal when it came to apply the Enlarged Board’s answers to the questions: T 116/18 Sumitomo/Insecticide compositions, unreported, 28 July 2023. The Board of Appeal began at [11.1] by expressing the view that the Enlarged Board’s focus in [93] on the application as filed and the filing date “is intended to prevent the filing of applications directed purely to speculative (armchair) inventions made only after the filing”. It reiterated this point at [11.8]. I note that this chimes with the reasoning of Lord Sumption in Warner-Lambert at [22] and [37 (third)].

At [11.10] the Board held that, for the first limb of the Enlarged Board’s test to be satisfied, “the purported technical effect together with the claimed subject-matter need only be conceptually comprised by the broadest technical teaching of the application as filed”, which did not mean that “said effect need … be literally disclosed in it by way of a positive verbal statement”.

At [11.11] the Board held that the second limb was satisfied unless the following question was answered in the affirmative: “would the skilled person, having the common general knowledge on the filing date in mind, and based on the application as filed, have legitimate reason to doubt that the purported technical effect can be achieved with the claimed subject-matter?”. The Board went on in [11.12] to reject the appellant’s argument that the second limb was only satisfied if “the application as filed contains experimental proof that the purported proof that purported technical effect is actually achieved with the claimed subject-matter”. I note that Warner-Lambert does not require experimental proof. On the contrary, plausibility may be demonstrated by experimental data falling short of proof or even by a priori reasoning unsupported by data: see Lord Sumption at [37 (fourth) and (sixth)]. The Board also held at [11.13] that it was not necessary for the application as filed to contain “a positive verbal statement about the purported technical effect”.

At [12] the Board expressly disagreed with this Court’s decision in Sandoz v BMS. It also partially disagreed with the Court of Appeal of the Hague in the case discussed below.

In AstraZeneca’s skeleton argument no reliance was placed upon any Board of Appeal decisions other than T 116/18. It was acknowledged in paragraph 88 that there were other Board of Appeal decisions that had sought to apply G 2/21 and that “they do not all take an identical approach to what G 2/21 means”. In oral argument counsel for AstraZeneca asserted that T 116/18 had been followed in “60-70” decisions, but he did not take the Court to any of these, and only made passing reference to two or three of them.

The Claimants relied upon T 314/20 Boehringer Ingelheim/Glucopyranosyl-substituted benzene derivative (unreported, a decision dated 19 November 2023, but the written reasons for which were only handed down on 20 December 2024) as demonstrating that the Boards of Appeal had not yet arrived at a settled interpretation of G 2/21. In this decision the Board agreed at [6.13.4] with T 116/18 that the purpose of the requirements laid down in G 2/21 was to prevent speculative applications for inventions which were only made after the filing date. At [6.13.5]-[6.13.11] the Board identified two issues with the reasoning in T 116/18. The first issue was that there was an inconsistency in the reasoning: the Board had stated that the Enlarged Board had not referred to any of the plausibility standards in its order, but instead adopted new requirements, and yet the Board had interpreted the second limb as imposing the ab initio implausibility standard. The second issue was that some of the Board’s inferences were not supported by the text of G 2/21. As the Board explained in [6.14], however, it did not need to reach a conclusion as to the correct test to be applied for the purpose of the case at hand.

In these circumstances it seems to me that it is not yet possible to say that the Boards of Appeal have adopted a settled view as to how the test in G 2/21 is to be interpreted and applied.

Turning to decisions of courts of other EPC Contracting States, AstraZeneca focussed on (translations of) decisions from the Netherlands, Spain and Switzerland, while the Claimants focussed on a decision from Norway. Before discussion of these decisions, it should be noted that many of the decisions cited in oral argument or referred to in the parties’ skeleton arguments concerned parallel decisions concerning the apixaban patent that was in issue in Sandoz v BMS. It is common ground that many other courts reached different conclusions to those reached by the UK courts. What matters for present purposes is the reasons for those differing conclusions. It is not uncommon for the courts of different EPC Contracting States to reach different conclusions on the same issue because of differences in the evidence and/or arguments. In the UK, an important aspect of the decisions is that Meade J made a finding, based on the expert evidence, that the skilled team would be aware of other potential reasons for apixaban having been prepared in greater quantities compared to other examples than the reason contended for by BMS, namely that apixaban was the most promising candidate of those exemplified in the application. This evidence does not appear to have featured in a number of the other decisions. There were also differences in the arguments. Finally, the reasoning of the courts differed from one another.

Of the Dutch decisions cited, it is sufficient to refer to those in the first case, which concerned an application for a preliminary injunction against a number of generic companies from marketing apixaban. The District Court of The Hague refused the injunction on the ground that the patent was probably invalid. The Court of Appeal disagreed and granted the injunction: [NL:GHDHA:2023:1593]. The Court considered G 2/21 at [6.5]-[6.13]. The generics argued that a stated technical effect may only be invoked if the skilled person “already understands from the patent application that the stated effect is actually achieved with the invention and the problem is actually solved, at least that this is made plausible”. The Court rejected this argument ([6.5]). The Court interpreted “would derive” in paragraph 2 of the Enlarged Board’s order as having the same meaning as the word “derivable” used in [72] ([6.6]-[6.7]). It noted that the Enlarged Board had differentiated between inventive step and sufficiency, and held that it would be incompatible with this to interpret G 2/21 as requiring that the alleged effect had been plausible when assessing inventive step ([6.8]-[6.9]). It agreed with BMS that the technical teaching of a patent “must be understood as ‘what is taught to the [skilled person] about how the technical problem can be solved with technical resources’” [6.10]. As can be seen from what the Court went on to say in [6.18] and [6.21.3] about the fact that the application disclosed a test with which the skilled person could easily determine that apixaban had a favourable Ki value, this amounts to an acceptance of the argument rejected by Lord Sumption in Warner-Lambert at [53] that it is enough for the application to encourage the skilled person to carry out simple tests identified in the application to confirm the efficacy of the claimed product if carrying out such tests would indeed show that the product is likely to be efficacious. That argument was not revived in Sandoz v BMS.

The Court then said at [6.12]:

“Contrary to what [the generics] argue, this interpretation of G 2/21 by the Court of Appeal does not lead to a free pass for speculative patents. After all, the fact that protection is granted based in a purely speculative patent for an invention that is only made afterward is prevented because it is required that the technical effect is already included by the technical teaching of the application and embodies the same invention that is disclosed therein. Moreover, it is established that EP 415 does not concern a speculative patent. BMS undisputedly argued that the inventors had already determined the favourable affinity and selectivity of apixaban experimentally prior to the submission of the application.”

The first two sentences appear to accept that speculative patents are not permitted by G 2/21. Counsel for AstraZeneca did not defend the reasoning in the second two sentences, which acquit BMS of speculative patenting on the basis that it had positive experimental data in respect of apixaban which it did not include in the patent application. This was not something relied upon by BMS in the English litigation, but it is also a point mentioned in a number of the other decisions. I cannot agree with this for the reasons given in paragraph 13 above. If reliance by the applicant upon secret data is ruled out, as it should be, then I remain of the view that I expressed in Sandoz v BMS at [93] that it is not possible to determine whether a claimed invention is speculative other than by assessing whether the application read in the light of the common general knowledge makes it plausible. As Lord Sumption said in Warner-Lambert at [37(third)], “[t]he disclosure of those grounds [sc. reasonable scientific grounds for expecting that it might well work] marks the difference between a speculation and a contribution to art”.

An appeal by the generic companies to the Dutch Supreme Court was rejected without reasons, from which it may be inferred that the Supreme Court agreed with the reasoning of the Advocate General in his opinion dated 20 December 2024 that the Court of Appeal had made no error justifying cassation. It is sufficient to note that the Advocate General expressed the view that G 2/21 “did not give an unambiguous answer” to the question of the extent to which the technical effect must have made plausible ([4.7]) and “does not solve all problems with plausibility inventiveness” ([4.11]) and that each case turned on its facts ([4.10] and [5.26]). Again, the Advocate General relied on what BMS had actually done before filing the patent, and not what was included in the patent, as an answer to the charge of speculative patenting ([5.8] and [5.14]).

Turning to Spain, AstraZeneca relied upon a decision of the Supreme Court concerning apixaban: Judgment 635/2015 dated 24 April 2025. It is common ground that that Court followed T 116/18 and applied an ab initio implausibility test. It does not appear that the Court was made aware of T 314/20. Furthermore, in concluding that the ab initio implausibility test was satisfied, the Court considered it significant that the ability of apixaban to inhibit Factor Xa, and its selectivity, “could be verified with tests described in the application”. The Court noted that the Court of Appeal of the Hague had reached the same conclusion concerning the application of the G 2/21 test to the apixaban patent.

As for Switzerland, AstraZeneca relied upon a decision of the Swiss Federal Patent Court concerning apixaban in Mepha Pharma AG v Bristol-Myers Squibb Holdings Ireland Unlimited Co dated 5 March 2024. As the Claimants pointed out, however, the Court held at [39] that:

“If the criterion that the claimed technical effect is considered by the skilled person to be embodied in the invention as originally disclosed is understood as the Boards of Appeal did in decision T 116/18 and T 1989/19 [a decision not cited to this Court that the Swiss Court stated had followed T 116/18], namely that this criterion is fulfilled if the skilled person has no reason to doubt that the invention achieves the technical effect, speculative applications are not prevented. In the case of a technical effect claimed in the original application without any basis, there are no reasons for its occurrence, but often also no reasons against its occurrence – especially since, in practically significant cases, it has subsequently been found that the invention achieves the claimed effect.”

This is a clear, and to my mind convincing, rejection of the ab initio implausibility standard.

The Court went on later in the same paragraph to interpret the second limb of the test in G 2/21 as follows (footnotes omitted):

“The second criterion is fulfilled if the technical effect is obvious to a person skilled in the art, taking into account the general knowledge at the time of filing, based on the application as filed. Experimental data or an express statement on the technical effect are not necessary for this purpose. However, since the purpose of derivability is to prevent abusive speculative applications, the hurdle must not be set too high, otherwise inventions that are worthy of protection will not be protected.”

I presume that by “obvious”, the Court meant “apparent”. In determining that this test was satisfied in the instant case, the Court held that it would not be unreasonable to expect the skilled person to determine the inhibition constants of all 110 examples disclosed in the application, and that when they did so they would find that apixaban had a Ki value of 0.08 nm and therefore conclude that apixaban was a potent and effective Factor Xa inhibitor ([42]). This reasoning is similar to that of the Dutch and Spanish courts, but it makes explicit the extent of the burden that this approach places on the public to carry out experiments that the patentee has not done (or if it has done them, not disclosed the results of) if the test is applied, as it should be, to the application. (If the invention is only plausible because of information in the granted patent which is not present in the application, then the patent is invalid for added matter: see Sandoz v BMS at [3] and [53].)

Inspired by the passage from the Swiss Court’s decision that I have quoted in paragraph 123 above, counsel for AstraZeneca argued that speculative patent applications were unobjectionable, and that it was only abusive speculative patents that were objectionable. I am unable to accept this submission for two reasons. First, it does not seem to me to supported by the decision. The Swiss Court accepted that “so-called ‘armchair’ inventions should not be rewarded with exclusive rights” ([39]). Its point was that it was important not to set the hurdle against speculative applications too high (a point which Lord Sumption recognised in Warner-Lambert at [36] and [37(third)]). As we have seen, the Board of Appeal in T 116/18 and the Court of Appeal of The Hague agreed that speculative applications should not be permitted. Secondly, I am unable to see what distinguishes an abusive speculative application from a non-abusive speculative application. Certainly, counsel for AstraZeneca was unable to articulate any workable criterion for distinguishing the two.

The Norwegian Court of Appeal in a decision concerning apixaban dated 3 March 2024 expressly rejected BMS’s argument that G 2/21 had “lowered the requirement for plausibility”. On the contrary, it held that:

“The decisive factor for whether a technical effect can be built upon in the assessment of inventive is whether at the time of application, based on the BMS application in combination with the general subject knowledge, it appeared credible – i.e. plausible – to the expert team that this effect would be achieved.”

On the facts, the Court considered that this test was satisfied. The reasoning was largely based on structural considerations rejected by Meade J on the evidence before him.

The conclusion I draw from this brief review of the case law of courts in other EPC Contracting States is that we are some way from unanimity as to what test G 2/21 lays down and how to apply it.There appears to be an emerging consensus that speculative patents are not permitted, but little agreement as to what constitutes a speculative patent. 

Accordingly, neither the jurisprudence of the Boards of Appeal nor the case law of courts of other Contracting States justifies a departure from Sandoz v BMS any more than G 2/21 itself does.

Even if AstraZeneca were to prevail on the law, and to establish that the correct standard to be applied to the assessment of inventive step was the ab initio implausibility, or no legitimate reason to doubt, standard, the Claimants contend that the Patent fails to satisfy even that standard. In support of this contention the Claimants rely, first, on the judge’s finding at [251] that the four Tanabe Seiyaku papers cited in WO 128 “would give the skilled team reason to doubt whether dapagliflozin would be an SGLT2 inhibitor”, and secondly, the judge’s finding in [259] that the absence of information about the EC₅₀ of dapagliflozin means that the skilled team cannot reasonably predict any useful effect on diabetes.

In my judgment the answer to the second contention is straightforward: there is nothing in the judge’s findings to suggest that the absence of information about the EC₅₀ of dapagliflozin would in itself give the skilled team a reason to doubt the efficacy of dapagliflozin to treat diabetes if they would not otherwise doubt the efficacy of dapagliflozin.

AstraZeneca advances two answers to the first contention in the alternative. The first answer is that there is an apparent inconsistency between the judge’s reasoning at [247]-[251] and what he said in [252] and [262]. At [248] the judge found that the skilled team reading the Patent would also read WO 128, and at [249] he found that they would also read the Tanabe Seiyaku papers. It is on this basis that he made his finding in [251]. Despite this, the judge went on to say in [252] that “it is not necessary to consider whether it would be permissible to rely on information in these cross-referenced documents to establish or support a case of plausibility applying … Warner-Lambert”. Moreover, in [262] he declined to decide whether the Patent satisfies a “legitimate reason to doubt” test on the ground that his findings of fact should be sufficient for a higher court to apply whatever test is appropriate. AstraZeneca contends that the better reading of the judgment is that the judge did not in fact make a finding about the effect of the Tanabe Seiyaku papers, since he did not decide whether they could be relied on. The Claimants dispute this.

In my view the better reading of the judgment is that the judge made findings at [248]-[251] that the skilled team would follow the cross-references and as to the conclusion they would draw from the Tanabe Seiyaku papers. In [252] he pointed out that this did not assist AstraZeneca. He left open the question of whether it was legally permissible for AstraZeneca to rely upon such information to establish or support plausibility. Although he did not spell out his reasons for leaving that open, it seems to me that it is probable that he considered that it was debatable applying Warner-Lambert whether it was permissible to go outside the four corners of the patent read together with the common general knowledge. By parity of reasoning, and having regard to [262], it seems to me that he also left open the question whether it was legally permissible for the Claimants to rely upon such information to establish or support ab initio implausibility.

Although the judge left those questions open, I consider that the logic of the judge’s reasoning applies whatever the test is. I see no reason why the skilled person or team should be confined to the four corners of the patent if it contains cross-references to other documents which they would follow up for the purposes of deciding whether the claims made were plausible or implausible. For the reasons the judge gave at [248] the skilled team reading the Patent would read WO 128 for that purpose, and for the reasons he gave in [249] they would also read the Tanabe Seiyaku papers. Having done so, they would reach the conclusion he set out in [251]. It follows that the skilled team would have legitimate reason to doubt that dapagliflozin is an SGLT2 inhibitor. That in turn means that they would have legitimate reason to doubt that it would be useful for the treatment of diabetes.

That takes me to AstraZeneca’s second answer to this point. If necessary, AstraZeneca contends by ground 6 that the judge erred (i) in taking the Tanabe Seiyaku papers into account and (ii) in finding that they would give the skilled team reason to doubt whether dapagliflozin would be an SGLT2 inhibitor. The difficulty that AstraZeneca faces with this ground of appeal is that it is a challenge to findings of fact made by the judge after hearing expert evidence.

AstraZeneca nevertheless submits that these findings were not open to the judge on the evidence. AstraZeneca argues that finding (i) was not open to the judge because none of the experts gave evidence that the skilled team reading the Patent would cross-refer first to WO 128 and then to the Tanabe Seiyaku papers. AstraZeneca argues that finding (ii) was not open to the judge because the upshot of Prof Potter’s evidence was that what was said about the compound called T-1095a “cannot inform you positively or negatively” about dapagliflozin.

I do not accept these arguments. In the case of finding (i), this was open to the judge on the evidence he referred to in [248] and [249]. In the case of finding (ii), it is fair to note that the judge did not expressly refer to the evidence of any of the experts in [251]. As counsel for the Claimants pointed out, however, Prof Potter was a cross-examined on this topic over 42 pages of transcript, and his evidence cannot be reduced to the soundbite relied upon by AstraZeneca. Furthermore, the judge had the advantage of being able to read the papers and see the experts cross-examined on them. This Court was not even provided with copies of the papers. This is island-hopping with a vengeance. Thus AstraZeneca has not demonstrated that the finding was not open to the judge. On the contrary, it appears to be supported by the abstract from the Hongu et al paper which the judge quoted at [250].

It must be assumed for the purposes of considering grounds 7 and 8 that at least claim 2 of the Patent is free from objection on plausibility grounds.

Ground 7 is that the judge was wrong in his approach to the issue of arbitrary selection. Ground 8 is that the judge failed properly to assess the contribution of WO 128 when considering whether the Patent made a technical contribution. Although counsel for AstraZeneca argued these two grounds together, it is important to distinguish between them.

The starting point for ground 7 is the final sentence of [59]. For convenience I will set this out again:

“If the patent claims a compound selected from a previously disclosed genus of compounds which are said to have a particular property, then that requirement is not satisfied if the compound does not in fact have some different or improved property compared to those previously individually disclosed (a new effect or an increase in an effect), or the patent does not make such improved property plausible.”

AstraZeneca contends that this is wrong in law, and that the judge ought to have held that, where the prior art merely asserted that the genus had a given advantage or property without rendering it plausible, then a subsequent patent may make a technical contribution by making the same assertion in respect of a selected compound and rendering it plausible. The judge rejected this argument at [271], but AstraZeneca argues that he was in error. I disagree. As the case law discussed by the judge makes clear, a selection from the prior-disclosed genus is only inventive if the selection makes a technical contribution because the selected compound in fact has some useful property which means that the selection is a technical advance. Mere plausibility is not enough for this purpose.

AstraZeneca also argued in its grounds of appeal that the technical contribution of the Patent was not merely that it made a plausible claim that dapagliflozin would be useful for the treatment of diabetes, but rather it made a claim to that effect which turned out to be true. This argument was rightly not pursued in AstraZeneca’s skeleton argument or oral submissions, because it is contrary to the way AstraZeneca put its case at trial as the judge explained at [266]-[268].

Given that I have rejected ground 7, success on ground 8 would not suffice for AstraZeneca to overturn the judge’s conclusion on arbitrary selection. I shall nevertheless address it. AstraZeneca argues that the judge went wrong because he compared what the Patent contributed to what WO 128 said, whereas he should have compared what the Patent contributed with what WO 128 contributed. AstraZeneca does not dispute that the relevant wording in the Patent is the same as that in WO 128, but it nevertheless contends that the quality of the disclosures is quite different. WO 128 makes an assertion about a class of compounds which it is common ground would at its narrowest (those covered by Formula IB) include millions of compounds, whereas the Patent is confined to dapagliflozin. AstraZeneca relies upon the principle, which is not in dispute, that it is easier to show a claim of narrow scope is plausible than a claim of broad scope.

In my judgment the judge was correct to reject this argument for the reasons he gave at [272]-[279] (see paragraphs 80-81 above). I would only add two points. First, counsel for AstraZeneca sought to attack the judge’s analysis of the expert evidence, but the judge’s findings are amply supported by that evidence. Secondly, counsel for the Claimants made a striking submission in support of the judge’s reasoning. Given the disclosure in WO 128, BMS could have filed 50 patent applications for compounds falling within Formula IB, including an application in respect of the compound of Example 12, and subsequently abandoned all 50 other than the one for dapagliflozin. On AstraZeneca’s argument, none of those applications would have been objectionable on the ground of arbitrary selection. What this submission highlights is the complete absence from the Patent of any basis for distinguishing between dapagliflozin and other compounds falling within Formula IB of WO 128 such as the compound of Example 12.

For the reasons given above I would dismiss the appeal. In summary, the judge correctly interpreted the Patent. Applying Sandoz v BMS, as he was bound to do, he correctly concluded that the Patent did not make it plausible that dapagliflozin would be useful for the treatment of diabetes, and therefore both claims 2 and 15 were invalid for both lack of inventive step and insufficient disclosure. Even if the standard applicable when considering inventive step for claim 2 is whether the skilled team would have legitimate reason to doubt that dapagliflozin would be useful for the treatment of diabetes, on the judge’s findings the skilled team would have legitimate reason to doubt this. Turning to arbitrary selection, the judge was correct to hold that the Patent claims a compound arbitrarily selected from the genus disclosed in Formula IB of WO 128 because the Patent makes no technical contribution compared to WO 128: the disclosure with respect to the assay is identical, and AstraZeneca does not even argue for this purpose that dapagliflozin is in fact superior in any of its properties to any of the compounds disclosed in WO 128, and in particular the compound of Example 12.

The judge made it clear that he reached his decision with regret given that dapagliflozin had proved to be successful, but as he rightly said at [281] “later findings about dapagliflozin do not enter the picture and my task has been to assess the validity of the Patent based on its disclosure”. The judge went on to observe that it was not for him to speculate about why the Patent was drafted in the way that it was, and in particular why it contained no data about the performance of dapagliflozin. It may be no coincidence, however, that the application for the Patent was filed by BMS at around the same time as the apixaban application. As discussed above, in the apixaban litigation courts of other EPC Contracting States have accepted that, at the time of filing the application for that patent, BMS had data in respect of apixaban which it did not include in the application. There is no evidence before this Court as to whether the same was true of dapagliflozin and the application for the Patent. Even if that were the case, I have explained above why applicants for patents should not be permitted to rely upon data which they choose to keep secret when filing their applications.