VPA treatment results in a hyperacetylation of histone H3 and H4
To create a more comprehensive picture of the dynamic histone code, we report the RAb along with the peptidoform-centric data, i.e. the measured peptidoforms with their combinatorial hPTMs (Fig. 3). Figure 3a depicts the RAb of all hPTMs competing for nine different acetylated residues. RAb estimates the percentage of the chromatin that is occupied by a given hPTM at a given residue. Importantly, we are able to detect very significant changes in very low hPTM levels occupying below 1% of the chromatin (e.g. Figure 3a: X-XIII). This will be highly relevant in the context of toxicity testing because localized at promotor or enhancer regions, these changes could induce considerable differences in expression of developmental mediators. The red lines clearly display an overall gain in acetylation levels as the VPA-concentration increases, confirming its action as an HDACi. In general, these acetylations replace other hPTMs, as is shown by e.g. the pronounced decrease in H3K9 methylation levels, which was already established for other HDACis (e.g. TSA) (Cozzolino et al. 2021). However, not all other hPTMs decline in response to a rise in acetylation. Most strikingly, dimethylation and trimethylation at H31K27 and dimethylation at H33K27 both rise with their respective acetylated forms, at the cost of monomethylated H31K27 and H33K27. As this is not a known direct enzymatic effect of HDACs, this implies that different histone writers are directly interacting or that cells react to the treatment (toxicity) by altering the activity of other histone writers. We recently showed that in both human and mouse ESCs, H3K27me2/3 is a gatekeeper of pluripotency and that the H4 N-tail is acetylated during differentiation (van Mierlo et al. 2019). Therefore, by chemically inducing H4 N-tail acetylation, the cell may increase H3K27 methylations to maintain pluripotency. These downstream or off-target effects will be important in future studies.
Figure 3b shows the different peptidoforms as they are measured by the LC-MS instrument after normalization for sample loading, without the subsequent RAb calculations applied, which can introduce certain biases (De Clerck et al. 2019b). When examining the peptidoforms containing acetylations (black arrowheads), the majority increases, especially at higher concentrations of HDACi. Noteworthy, some acetylations are not affected, possibly (i) because of a neighboring PTM blocking enzymatic interaction, (ii) because these sites are not a substrate for the histone acetyl transferase (HAT) or HDAC, or (iii) because they are an intermediate form that is modified further into a hyperacetylated form at higher VPA concentrations (e.g. H4(4–17): Mono-Ac). Interestingly, an additional internal validation of the performance of the workflow is the opposing trend exhibited by H31K36me2 (Fig. 3b highlighted by red arrowhead) compared to H31K27me3, a recently described direct interaction discovered by using advanced computational models (Alabert et al. 2020).
In conclusion, this data illustrates the applicability of our untargeted workflow to simultaneously map changes in many different hPTMs.
To illustrate the scalability of the workflow, hESCs were incubated with four different concentrations of in total ten different compounds (VPA included) in quadruplicate each: i) drugs with a known effect on the histone code (BIX, DZNep, TSA and VPA), ii) drugs that are classified as highly embryotoxic by the ECVAM (MTX and RA), iii) a drug that is classified as non-embryotoxic by ECVAM (PenG), and iv) common substances of abuse with a presumed developmental toxicity (caffeine, ethanol and nicotine). The impact of some of these compounds on specific histone marks has been studied in the past with Chromatin immunoprecipitation sequencing (ChIP-seq) (Pal-Bhadra et al. 2007; Ping et al. 2014; Subbanna et al. 2013; Urvalek and Gudas 2014). However, ChIP-seq can only obtain information about specific locations in the genome. In contrast, our workflow does not focus on a specific modification site, and therefore can be used complementary to detect targets of interest while also taking combinatorial hPTMs into account.
Again, all cells were monitored for pluripotency and cell death using flow cytometry and RT-qPCR. No other treatment then VPA led to loss of pluripotency of the hESCs within the timeframe of the experiment, i.e. 24 hours, as none of the lineage markers significantly changed as a function of concentration for the other compounds (supplementary Data S2). Nevertheless, due to the highly toxic nature of TSA, an excessive number of cells were dying during incubation at the highest concentration. This was observed by cell detachment from the vitronectin plate, and by flow cytometry after harvest (Supplementary Figure S2 and Supplementary Data S5). Only 71.6% of the remaining cells were still alive and available for harvest after treatment with 100nM TSA. This made further histone analysis irrelevant and therefore only three remaining concentrations for TSA were subjected to histone analysis.
Figure 4 depicts the fold changes for the increasing concentrations against negative control samples (hESCs incubated in H2O or DMSO, depending on the solvent involved) for a set of peptidoforms of H3 and H4.
Compounds with a known effect on the histone code
BIX and DZNep are histone methyltransferase inhibitors (HMTis) and TSA and VPA are known HDACis. Note that inhibiting a methyltransferase will reduce methylation, while inhibiting deacetylases will increase acetylation. Indeed, HMTis are the only compounds in which the trimethylation of H31K27 was not observed, in line with the fact that DZNep is capable of inhibiting EZH2, a writer of H3K27me3. Furthermore, DZNep has a known effect on the methylation of H4K20 (Miranda et al. 2009), which was observed as well. Still, a more general inhibition of both repressive and active histone methylation marks was observed as illustrated by the notable decrease in H3K9me2, H3K79me2 and H4K20me2. Also for BIX, an inhibitor of a G9a histone methyltransferase, our findings are in strong agreement with the literature, since the G9a enzyme is responsible for the methylation of H3K9 (Ciechomska et al. 2016). Yet a more global effect in methylation is also visible here, as seen by a decrease in H3K27Me2, H3K9Me2 and H3K9Me3. For TSA, a pan-HDAC inhibitor originally known as an antifungal antibiotic, the results are very similar to those already discussed for VPA, with an overall distinct increase in acetylation levels, along with a rise in H31K27me2, H33K27me2, and H31K27me3 and a decrease in methylation of H31K9 and H31K36me2. Most of these findings are in line with the recently described effects of TSA on mouse ESCs (Cozzolino et al. 2021). Moreover, we recently described for the first time that the histone code changes in a very similar way between mouse and human ESC during differentiation, making mouse a potential model for developmental toxicoepigenetics (De Clerck et al. 2019a).
Compounds from the ECVAM-classification
The effect of MTX on hPTMs has, to the best of our knowledge, never been investigated, yet we show that this strong embryotoxic compound displays a very prominent and non-coherent dysregulation of the hPTMs. Some hPTMs do not show a concentration-dependence and are heavily affected, even at the lowest concentration (e.g. Figure 4: H31(73–83): K79[Me2] and H33(27–40): K27[Me2]). Therefore, our study could provide a steppingstone to explore the histone fingerprint of MTX more profoundly, for example, by incorporating lower concentrations of MTX or in a time-lapse experimental design. Surprisingly, for RA, another strong embryotoxic compound used for the treatment of acne and acute promyelocytic leukemia, the induced changes are much more subtle within the investigated timeframe. The most prominent change is a decrease in H3K27me3, the hallmark of pluripotency, which is in agreement with the ability of RA to induce differentiation in ESCs (De Angelis et al. 2018). The fact that no other lineage specification genes or Oct4 protein change was observed (supplementary Data S2) is in line with the epigenetic role of H3K27me3, which precedes expressional differences. Noteworthy, the hPTM-profile of RA is more similar to that of PenG than to MTX, suggesting that the embryotoxicity of RA is either i) not mediated by hPTMs, ii) only emerging at higher concentrations, or iii) a long-term effect that was not sampled in the experimental design. Finally, PenG, a broad-spectrum, beta-lactam antibiotic, was included as a negative control, i.e. not embryotoxic by the ECVAM. No strongly pronounced changes, except for a concentration-dependent effect on K79 monomethylation and a very subtle increase in H4 N-tail acetylation, are observed for PenG in this experimental design.
Compounds of abuse
Finally, the compounds of abuse displayed relatively moderate fluctuations in their histone signature. Still, caffeine exhibits the most pronounced pattern which, when directly matched, resembles that of MTX most closely (Figure 5). This is a finding of concern. Currently, it is recommended by the World Health Organization (WHO) not to consume more than 300 mg of caffeine per day during pregnancy because excessive intake may be associated with growth restriction, decreased birth weight, preterm birth or stillbirth (Guilbert 2003; Sengpiel et al. 2013). Our data suggests that these toxic effects might be linked to changes induced in the histone code. Nevertheless, it should be noted that the metabolization of caffeine was not considered in this experiment. Next, we included ethanol because of its established negative impact during gestation, referred to as fetal alcohol spectrum disorders. Overall, ethanol displays very subtle fold changes, however it does seem to mirror PenG, our negative control in terms of embryotoxicity (Figure 5). This suggests that the effect of a one-time intake of ethanol has only a limited influence on the histone code. With several contradictory findings on the effect of ethanol on specific histone marks published earlier, we conclude that more accurate quantification and robust statistical data analysis strategies are required to resolve these very subtle changes in the MS data (Lo and Zhou 2014; Pal-Bhadra et al. 2007; Subbanna et al. 2013). This also holds for nicotine, the addictive compound in cigarette smoke. Smoking is known for its negative impact on pregnancy, e.g. increasing risk of preterm birth, lower birth weight, miscarriage, birth defects, and Sudden Infant Death syndrome (Wickstrom 2007). Whereas the toxicity of nicotine has been widely studied, its impact on hPTMs has only been studied in differentiated tissues (Chase and Sharma 2013). Our toxicoepigenetic workflow shows that the hPTM-changes for nicotine are so subtle that it is very conceivable that a one-day intake of nicotine does not affect the hPTMs in stem cells. Again, advanced data analysis strategies need to be developed to make this conclusion more founded.
To date, little is known about the effects of different compounds on the hPTM-landscape. Yet, our comprehensive overview of the hPTM changes induced by ten compounds in stem cells shows that most compounds have a (subtle) effect on the histone code.
Our study is the ideal steppingstone to extend the knowledge on this form of epigenetic toxicity in light of developmental toxicity. This can be done by i) including other compounds of interest, ii) adjustment of dose, iii) adjustment of incubation time or the use of time-lapse experimental designs and, iv) developing more advanced statistical methods and algorithms to cluster compounds to facilitate the decision-making toolbox. Moreover, the applicability of our workflow goes far beyond developmental toxicity. Firstly, other forms of toxicity can also be investigated depending on the cell line used, e.g. hepato-and nephrotoxicity by using liver and kidney cells respectively. Secondly, this study is not only important in the context of toxicoepigenetics but is also a promising tool in the field of pharmacoepigenetics. As these epigenetic modifications are interesting targets due to their dynamic and reversible character, the development of epidrugs is gaining momentum. Especially in oncology, the use of epidrugs is on the rise and our workflow may contribute to discovering or elucidating the mechanism of action of these drugs (Miranda Furtado et al. 2019; Montalvo-Casimiro et al. 2020). Moreover, personalized medicine is receiving growing attention and this study can contribute to this as well (Rasool et al. 2015). For example, it is possible to determine whether a patient exhibits a particular hPTM characteristic on which the drug will act, thereby predicting whether or not the treatment is likely to succeed. Finally, the scope of this study can be extended outside the pharmaceutical context including applications for environmental toxicity and food safety.
However, to make the results of this workflow easier to interpret, more reliable and consequently easier to implement, we are still working on some improvements both in terms of acquisition and data analysis. For instance, with our current LC-MS/MS settings, it is difficult to acquire modified forms of H3K4. There are two reasons for this: (i) this PTM site is located on a small tryptic peptide, that consequently elutes early, making it difficult to analyze and, (ii) our mobile phase contains DMSO, which improves ionization, but also causes charge state reduction. Therefore, the H3K4 peptide occurs mostly as a singly charged ion (Hahne et al. 2013). Nevertheless, this modification site can be of interest as methylation of H3K4 is associated with active transcription. Consequently, besides optimization of the LC gradient, acquisition parameters can be adjusted to also target singly charged precursors or DMSO can be removed from the mobile phase to include modifications of H3K4 in the future. These improvements in LC-MS/MS settings should also result in a better separation of other peptides, which in turn will allow more accurate quantification, so that differences will become even more apparent. Note that the effect of withdrawing DMSO on the other histone peptides should be assessed as well, since it is known that doubly charged peptides are best annotated as they mostly generate singly charged fragments. Furthermore, including data-independent acquisition technologies like (Scanning)SWATH will result in an improved quantification and will be a stepping stone in the transition towards a multiple or parallel reaction monitoring, respectively MRM and PRM, assay (De Clerck et al. 2019b). When focusing on data-analysis, we already mentioned that caution is always required when reporting RAbs (De Clerck et al. 2019b). Depending on the peptidoforms used for the calculation in combination with the ionization effects, RAbs can lead to a confusing and even misleading form of reporting. Therefore, we are working on more advanced statistical approaches that could contribute to better reporting and consequently a better understanding of the outcomes.
In conclusion, we demonstrated that with our workflow toxicoepigenetic screening on histones is feasible and will yield very rich data, for which more streamlined interpretation tools are yet to be developed. Integration of this epigenetic information into the field of toxicology is a promising addition that offers an opportunity to gain novel insights into toxicological phenomena (McCullough and Dolinoy 2018). We envision a future wherein 100–200 histone peptidoforms are brought together in a single MRM or PRM assay that runs in < 10 minutes per sample, enabling 6 samples per hour or nearly 150 samples per day per instrument, which get automatically analyzed to create a user-friendly report. Storing all results in a central database will finally allow to cluster novel compounds with other, known toxicoepigenetic effects, classifying them according to potential toxicity level in a given targeted cell type. As a result, this proof-of-concept to develop a screening assay can contribute to the (safe) development of drugs as well as to the field of environmental toxicity and food safety.