Off-target profiling of tofacitinib and baricitinib by machine learning : a focus on thrombosis and viral infection

Objectives: There are no clear on-target mechanisms that explain the increased risk for thrombosis and viral infection or reactivation associated with Janus kinase (JAK) inhibitors. We aimed to identify and validate off-target binding effects of the JAK inhibitors baricitinib and tofacitinib using computational and experimental methods. Methods: Potential biological targets of baricitinib and tofacitinib were predicted using two established computational methods. Targets related to thrombosis or viral infection/reactivation were experimentally validated using biochemical and cell-based in vitro assays. Results: Overall, 98 targets were predicted by the computational methods (baricitinib n=41; tofacitinib n=58), of which 17 drug-target pairs were related to thrombosis (n=10) or viral infection/reactivation (n=7), and 11 were commercially available for in vitro analysis. Inhibitory activity was identified in vitro for four drug-target pairs – two related to thrombosis in the micromolar range (phosphodiesterase 10A [baricitinib], transient receptor potential cation channel subfamily M subtype 6 [tofacitinib]) and two related to viral infection/reactivation in the nanomolar range (Serine/threonine protein kinase N2 [baricitinib, tofacitinib]). Conclusions: Previously unknown off-target interactions for the two JAK inhibitors were identified. The proposed pharmacological off-target effects include attenuation of pulmonary vascular remodeling, modulation of Hepatitis C viral response and hypomagnesemia. Off-target effects related to an increased risk of thrombosis or viral infection/reactivation for baricitinib and tofacitinib were not identified. Further clinical and experimental research is required to explain the observed thrombosis and viral infection/reactivation risk.

. Chemical structure of baricitinib and tofacitinib, the first JAK inhibitors approved in the USA and Europe to treat rheumatoid arthritis.
Safety concerns associated with JAK inhibitors, such as the increased risk for thrombosis and viral infection or reactivation have emerged worldwide and boxed warnings are included on all approved JAK inhibitors used to treat inflammatory conditions.(4-7) While a dose-response effect was observed in the risk of thrombosis in clinical trials of both baricitinib and tofacitinib, there is no clear mechanism associated to the pharmacological target that could explain the risk of thrombosis associated with JAK inhibitors. Thus, the increased risk of these safety concerns is heavily debated.
It is well established that unintended off-target activity may interfere in multiple biological processes, inducing unwanted side effects.(8) In this context, computational approaches, such as machine learning, can be used to predict the potential for an approved drug to interact with off-targets and identify potential safety-related concerns.(9,10) For example, previously unknown drug-target interactions for the approved compound Celecoxib were identified, supporting the biological plausibility of reported cardiovascular adverse drug events.(11) Off-target profiling is frequently used to identify candidate drugs for repurposing. For example, computational studies identified baricitinib as a promising JAK inhibitor for repurposing in patients with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2 or COVID- 19).(12,13) Baricitinib was considered to be a therapeutic option based on the high affinity for AP-2 associated protein kinase 1 (AAK1), which is key in regulating viral endocytosis(12) and its inhibition may reduce the ability of the virus to infect lung cells. (14) In light of the currently unexplained thrombotic and viral infection risk, and the previously observed off-target binding potential of baricitinib, we sought to identify if the thrombosis and viral infection/reactivation risk may be a result of an off-target effect.
We therefore aimed to leverage computational methods to identify unexplored drugprotein interactions for baricitinib and tofacitinib in vitro.

Macromolecular target prediction and selection.
Macromolecular targets of baricitinib and tofacitinib were predicted using two machine learning approaches, Target Inference Generator (TIGER v. 19.07)(11) and SOM-based Prediction of Drug Equivalence Relationships (SPiDER)(15). TIGER and SPiDER leverage self-organizing maps (16) for target prediction (see online supplementary material 1).
Targets with statistically meaningful predictions from SPiDER (p<0.05) and/or TIGER (score>1) were selected for in vitro testing if they were considered to have a potential influence in thrombosis and viral infection/reactivation.

In vitro characterization.
Baricitinib (99.97% purity) and tofacitinib (99.96% purity) compounds were purchased from MedChem Express LLC (New Jersey, www.medchemexpress.com). In vitro characterization was performed on a fee-for-service basis at Eurofins (www.eurofins.com), if the assay was commercially available. Baricitinib and tofacitinib were tested at a single concentration (30 µM) or multiple concentrations (30 µM highest concentration) with technical replicates, using biochemical assays (radioligand or enzymatic assays) or cell-based assays (see online supplementary material 2 for further details).

RESULTS:
The target prediction methods identified 40 potential targets for baricitinib and 58 for tofacitinib (online supplementary tables 1 and 2, respectively). From all predicted targets, five targets for baricitinib and five for tofacitinib were identified as being relevant for thrombosis (Table 1). For viral infection/reactivation (Table 2), four targets were identified for baricitinib and three for tofacitinib.  *Commercial assays were unavailable for TRPC6 or TRPC3, and therefore, these targets could not be validated. Instead, transient receptor potential cation channel subfamily M member 6 (TRPM6) was employed for the respective binding assays. Serine/threonine-protein kinase N2 (PKN2)* Thymidine kinase (HSV)

Tofacitinib
Exportin-1 (XPO1) Serine/threonine-protein kinase N2 (PKN2) Ubiquitin-conjugating enzyme E2 N (Ubc13) * PKN2 was included in the list of targets tested for baricitinib, which allowed us to make a direct comparison between tofacitinib and baricitinib inhibitory activity on this target.
Of the 98 predicted targets, a total of 11 drug-target interactions were experimentally validated based on the availability of fee-based in vitro testing services (Table 3). Among predicted targets, two members of the Transient Receptor Potential superfamily of calcium channels were suggested, namely short transient receptor potential channels 6 (TRPC6) and 3 (TRPC3). Commercial assays were unavailable for TRPC6 or TRPC3, and therefore, these targets could not be validated. Instead, transient receptor potential cation channel subfamily M member 6 (TRPM6) was employed for the respective binding assays.

DISCUSSION:
The confirmed drug-target interactions suggest an attenuation of pulmonary vascular remodeling (inhibition of PDE10A), modulation of Hepatitis C (HCV) viral response (inhibition of PKN2), and hypomagnesemia (inhibition of TRPM6). Therefore, we did not Rather, baricitinib might improve progressive pulmonary vascular remodeling.
This study further identified previously unknown off-target interactions of tofacitinib on the TRPM6, with moderate binding affinity. While our computational approach identified TRPC6 and TRPC3 as potential targets, we were unable to experimentally validate these targets due to lack of commercially-available in vitro assays. In addition to thrombosis, targets related to viral infection and viral reactivation were Outside of its role in viral suppression, PKN2 may play an essential role in various cellular processes, such as cellular proliferation, migration, and signaling pathways. (42)(43)(44) Moreover, PKN2 is involved in autoinflammatory disorders,(45) heart failure, (46) and it is a target of interest in cancer. (44,47,48) As concerns regarding the risk of malignancy and major adverse cardiovascular events (MACE) in patients treated with tofacitinib have been raised by the European Medicines Agency (EMA), it is important to consider the potential role of PKN2 inhibition. (49) However, in mice models, PKN2 activation was the cause of cardiac dysfunctions, (46) and therefore, the clinical impact of PKN2 inhibition is contradictory to the risk of cancer and MACE in RA patients.
Off-target profiling using computational approaches has been widely used to identify candidates for drug repurposing. (50,51) Indeed, JAK inhibitors were recently established as potential candidate therapies for SARS-CoV-2 based on in silico methods. (52)(53)(54) Our computational methods identified 98 drug-target predictions, and the preliminary in vitro results found inhibitory activity on several proteins other than the primary therapeutic target, thereby confirming baricitinib and tofacitinib as promiscuous drugs (55,56) and candidates for drug repurposing studies. For example, PDE10A inhibition has been primarily studied in psychiatric and neurological conditions, such as schizophrenia (57) and Huntington's disease, (58) and, to a lesser extent, in multiple peripheral pathological conditions(59,60) (e.g., osteogenic differentiation). Additionally, PDE10A inhibition by baricitinib is hypothesized to have a synergistic pharmacological effect in combination with other COVID-19 treatments (e.g., antiviral and corticosteroids drugs), due to the antifibrotic and anti-inflammatory effects of phosphodiesterase's inhibitors on the treatment of COVID-19 and its associated conditions (e.g., thrombosis, inflammation, and fibrosis).(61,62) Therefore, the confirmed PDE10A inhibition identified in this study supports the potential for baricitinib as a potential candidates outside of rheumatology.
Moreover, while TRPM6 was not initially predicted the moderate inhibitory active is worth investigating. TRPM6 inhibition is not fully elucidated, however, it is mainly involved in magnesium homeostasis in the intestine and kidney (63,64) and it has been demonstrated to have expression levels modulated by hormones such as estrogen (65) and angiotensin II, (66) immunosuppressant (67)  draw. Thus, we encourage researchers with access to the appropriate assays to validate these targets. Moreover, there might be additional targets of relevance that were not predicted by our computational tools. We also acknowledge that the activity of small molecule drugs using in vitro assays does not always translate into activity in the cellular environment. Thus, the results should still be interpreted with caution and treated as preliminary evidence for the off-target binding of baricitinib and tofacitinib.

CONCLUSION:
The combination of computational methods and experimental validation identified and characterized previously unknown off-targets of baricitinib and tofacitinib.
The confirmed target interactions suggest an attenuation of pulmonary vascular remodeling, modulation of HCV viral response, and hypomagnesemia, thus it does not endorse the hypothesis of elevated thrombosis or viral infections/reactivation risk explained by one (or more) drug-target interactions. Consequently, the current safety concerns may be due to underlying patient-specific factors (confounders) or to targets not detected by our computational pipeline, but as not all of the predicted targets related to thrombosis or viral infection/reactivation were experimentally validated further research is warranted. Finally, baricitinib and tofacitinib may be potential candidates for repurposing, as they were identified as drugs with promiscuous binding activity.