Cell Lysate Microarray for Mapping the Network of Genetic Regulators for Histone Marks

Protein, as the major executer for cell progresses and functions, its abundance and the level of post-translational modifications, are tightly monitored by regulators. Genetic perturbation could help us to understand the relationships between genes and protein functions. Herein, we developed a cell lysate microarray on kilo-conditions (CLICK) from 4,837 yeast knockout (YKO) strains and 322 temperature-sensitive mutant strains to explore the impact of the genome-wide interruption on certain protein. Taking histone marks as examples, a general workflow was established for the global identification of upstream regulators. Through a single CLICK array test, we obtained a series of regulators for H3K4me3 which covers most of the known regulators in Saccharomyces cerevisiae. We also noted that several group of proteins that are linked to negatively regulation of H3K4me3. Further, we discovered that Cab4p and Cab5p, two key enzymes of CoA biosynthesis, play central roles in histone acylation. Because of its general applicability, CLICK array could be easily adopted to rapid and global identification of upstream protein/enzyme(s) that regulate/modify the level of a protein or the posttranslational modification of a non-histone protein.


INTRODUCTION
Genetic perturbation has been widely employed to illuminate gene function in biological research. With the increasing pace of technology advancement in genome editing approach and sequencing, several large-scale genetic perturbation approaches have been developed to map the genetic element regulators in cell phenotypes such as cell viability (Blomen et al., 2015), cell differentiation (Jaitin et al., 2016) and cell immune response (Parnas et al., 2015), and etc. However, not all genetic perturbations resulted in obvious change of cell state. Protein, the major performer in cell processes, should be a better readout for genetic perturbation. However, genome-wide perturbation method to identify the network of genetic regulators on protein state, including its abundance and post-translational level, are still largely lacking. Recently, a method using random mutagenesis in haploid human cells is applied to protein measurements in individual cells (Brockmann et al., 2017). However, this method relies on fluorescence activated cell sorter (FACS) and deep sequencing which require professional and complicate operation. Thus, there is a need to develop a quick and efficient method for globally probing of the genetic regulators and the underlying network on protein state.
Histones are one of the most important and conserved proteins in eukaryotic cells.
They are subjected to many different types of post-translational modifications (PTMs), especially many novel PTMs discovered in the past few years  including a wide range of lysine acylations, such as butyrylation (Chen et al., 2007), crotonylation (Tan et al., 2011) and β-hydroxybutyrylation (Xie et al., 2016). Currently, there are more than 500 known histone marks . Many of these histone marks play important roles in the regulation of DNA-related processes such as replication and transcription. However, how and by what proteins these histone marks are regulated remains largely unknown. The proteins affecting the dynamics of histone marks include "writers", "erasers", "readers", and other regulators (Strahl and Allis, 2000). We define in this study all of these proteins as regulators of histone marks (RHMs), and classify them into two categories: positive-regulators or negative-4 / 33 regulators which perturbation could decrease or increase, respectively, the level of the corresponding histone mark.
There are several existing strategies for the discovery of RHMs. Some RHMs have been identified based on the chemical similarity of the particular modification. For example, since propionylation, butyrylation and crotonylation are structurally similar to acetylation, it was expected that the "writers" and "erasers" for acetylation (namely, histone acetyltransferases (HATs) and histone deacetylases (HDACs)), also regulate these other acylation states, and indeed, this was confirmed for crotonylation (Bao et al., 2014;Sabari et al., 2015). However, despite the hundreds of acylation modifications on histones, there are very few regulators have been identified to date.
We anticipated that this number of regulators is too low to be responsible for the tight and pervasive regulation of diverse histone acylation states, and so expected that many novel regulators remain to be discovered.
An alternate strategy for RHM discovery is to screen thousands of random or targeted mutant cells such as an entire collection of a deletion library in an unbiased way by immunoblotting (Krogan et al., 2003). Clearly however, it is not a trivial task to perform immunoblotting on thousands of samples. Recently, chromatin immunoprecipitation coupled with mass spectrometry (ChIP-MS) has been used to identify proteins bound at genomic regions associated with specific histones marks (Ji et al., 2015). However, this method relies on stable interactions between proteins and histones, which may not be suitable to identify indirect regulators that do not bind the histones. To accelerate functional studies of histone marks, an efficient strategy for the fast and global identification of regulators at an affordable cost could be transformative.
Saccharomyces cerevisiae is a well-studied eukaryotic model organism. There are approximate 1,000 human disease related genes that are evolutionary conserved from S. cerevisiae to human (Heinicke et al., 2007). Besides methylation and acetylation, a variety of other histone marks have also been identified on histones in S. cerevisiae, such as 2-hydroxyisobutyrylation (Dai et al., 2014), succinylation and malonylation (Xie 5 / 33 et al., 2012). S. cerevisiae is a powerful system for genetic and epigenetic studies with many available tools and resources, such as the yeast knockout (YKO) collections that carry precise start-to-stop deletions of ~6,000 open reading frames (Giaever and Nislow, 2014) and a collection of temperature-sensitive (TS) mutants spanning 497 different essential genes (Ben-Aroya et al., 2008) . Previously, the YKO collection has been successfully applied for the screening of histone PTM regulators by western blotting (Krogan et al., 2003) or dot blotting (Han et al., 2007). However, these strategies were significantly labor-intensive and time-consuming, limiting their general application for the discovery of histone regulators.
Microarrays, e.g. protein microarrays, are powerful tools for systematic analysis/discovery (Zhu et al., 2001). Notably, microarrays enable proteome-wide analysis while consuming tiny amount of samples in a short period of time. For example, the reverse phase protein array (RPPA) (Paweletz et al., 2001) contains hundreds to thousands of lysates from individual samples, e.g. tumors. By applying a set of antibodies for a specific pathway, the abundance of specific components of this pathway could be monitored quantitatively across many samples in a single test (Chan et al., 2004).
In this study, we applied the protein microarray technology to fabricate a cell lysate microarray on kilo-conditions (CLICK) with the haploid YKO collection and the TS mutant strains to prepare an array enabling the examination of ~5,000 single proteins as potential RHMs in a single experiment. Using antibodies specific for H3K4me3 and H3K36me3, we demonstrate that most of the known regulators for these two histone marks could indeed be discovered in a single round of the CLICK assay. Next, using antibodies against many specific acyl modifications, we found that Cab4p and Cab5p, the final two enzymes of CoA biosynthesis, play central roles in acylation on histones.
Since the CoA biosynthesis pathway is highly conserved, these RHMs may function to modulate histone acylation within eukaryotes more broadly. We anticipate that the CLICK array technology described here will become a generally applicable strategy for the systematical identification of RHMs and regulators of other proteins for both PTMs 6 / 33 and abundance.

Overview and construction of the CLICK array
The schematic diagram of the CLICK array is depicted in Figure 1A. The collection containing thousands of certain gene deletion or mutant are cultured in certain condition. Different cell lysates represent different states for certain protein. Taking the histone mark as an example, in some cell lysates with positive-RHMs (such as gene B) or negative-RHMs (such as gene D) genetic perturbation, the histone mark level could be decreased or increased, respectively. Then these cell lysates are deposited onto a microarray which will be probed with two different primary antibodies simultaneously: one that targets histones generically and another that is specific for a particular histone mark. Subsequent probing of this array with fluorescently-labeled secondary antibodies specific for the primary antibodies enables a direct measurement of the levels of the particular histone mark and histone from the same spot. The ratio of the levels of histone mark/histone are calculated for each strain and compared to that of the WT. Variations such as batch-to-batch and array-to-array differences are greatly reduced with this dual color labeling strategy, thereby ensuring the reliability of the measurements. The positive-RHMs (the red spots on the merged array) and the negative-RHMs (the green spots on the merged array) are readily identified according to the rank of the ratio of histone mark/histone levels. With these regulators interaction information, we could easily draw a upstream regulation network for certain protein or protein PTMs.
One key factor of the CLICK array is the antibody specificity. Unlike assays such as western blotting that can tolerate some cross-reactivity in an antibody since the other proteins to which it binds can be separated from the protein of interest, this is not possible with the CLICK array: the complete collection of proteins of a strain are all immobilized on a single spot, and so any cross-reactions may lead to a strong collective signal. Therefore, preliminary assays to screen for CLICK-grade antibodies were first 7 / 33 carried out. In particular, we examined the specificity of a given antibody by immunoblotting against the total lysates of WT strains (S. cerevisiae strains BY4741 and BY4742). A negative control consisting of a mutant histone protein in which the residue with the modification was mutated to alanine was included. That is, for example, H3K4A was included when testing an antibody for H3K4me3. An antibody was qualified for use in the CLICK array only if one major band at the expected size was present in the immunoblot, and at the same time no band or only very weak band was observed in the mutant sample. After screening a variety of modification-specific antibodies from different sources, we identified a set of antibodies that satisfied this criteria, including antibodies for H3K4me3, H4K16ac ( Figure 1B), H3K36me3, H4K8ac, H3K9butyryl and H3S10p ( Figure S1A). A similar examination of antibodies specific for histones H3 or H4 from a variety of manufactures, to be used as "loading controls" in the CLICK array, also identified several excellent candidates that only showed one major band in the total lysate from the WT strains ( Figure 1B).
To assure the specificity of a histone mark-specific antibody, we next carried out dot blotting, which is more similar to the CLICK array since all of the cellular proteins are retained in one single spot. The ratios of the histone mark and histone (H3 or H4) were then measured. We found that among the six antibodies identified as highly specific in the western blotting assay, five showed similar levels of specificity in the dot blotting assay ( Figure S1B): the antibody for H3K9butyryl was the sole candidate that failed in this assay ( Figure S1C).
To construct the CLICK array, 4,873 nonessential deletion mutants from the haploid YKO collection and 322 temperature-sensitive essential genes mutants were included.
To validate the quality of the resulting array, we probed it with a H3K4me3 specific rabbit polyclonal antibody and a H3 specific mouse monoclonal antibody, followed by

/ 33
Cy3-and Cy5-labeled secondary antibodies specific for rabbit and mouse, respectively ( Figure. 1C). As expected, most of the spots displayed yellow, indicating that the ratios of H3K4me3 to H3 were constant across many mutant strains. That is, the majority of genes did not affect the level of H3K4me3 ( Figure. 1C). Measuring the signal intensity of histone H3, the distributions of foreground and background signal intensity exhibited typical bell-shaped curves that were almost completely separate (Figure S2), indicating that the vast majority of the printed spots on the microarray contained substantial levels of histones.

Identification of the network of RHMs of H3K4me3 by the CLICK array
To test the applicability of the CLICK array, we first examined for RHMs of H3K4me3, one of the best-characterized histone marks. The CLICK array was probed with the H3K4me3-specific antibody, and the fold-change of modification-to-histone level was determined as: for yeast strain X, the fold-change is (histone mark signal-X / histone signal-X) / (histone mark signal-WT/ histone signal-WT), where the histone mark is H3K4me3 here, and WT is the BY4742 strain. The fold-changes for all strains on the CLICK array were then determined. Overall, we identified 21 positive-regulators and 61 negative-regulators (Figure 2A) and some of them were verified by western blot ( Figure 2B and 2C). To illustrate the relationships among these RHMs, we combined our data with the known regulators prior to this study and analyzed their interactions using STRING (Szklarczyk et al., 2017), where interacting proteins were grouped based on their GO classifications ( Figure 2D). We anticipated to re-discovery previous known RHMs and some new RHMs. As expected, even though several known regulators were absent on the CLICK array, our data had covered all known complexes which were reported to regulate H3K4me3, including the COMPASS (SWD3, BRE2, SWD1, SPP1 and SDC1) which mediates trimethylation of H3K4 (Dehe et al., 2006), Paf1C (Polymerase Associated Factors 1 complex) (CDC73) which was shown to be required for H3K4me3, H3K36me3 and H2BK123ub (Krogan et al., 2003;Wood et al., 2003), the Rad6p-Bre1p-Lge1p complex (RAD6 and LGE1) which is responsible for the ubiquitylation of 9 / 33 H2BK123 (Hwang et al., 2003), the BUR kinase complex (SGV1) (Laribee et al., 2005) and CCR4/NOT complex (POP2 and MOT2)   (Figure 2D).
Surprisingly, we found that knockout of ARD1, encoding a subunit of protein Nterminal acetyltransferase NatA, which is known to regulate trimethylation of H3K79 (Takahashi et al.,2011), caused a 60% reduction of H3K4me3 (Figure 2B), implicating ARD1 as a previously unknown mediator of K79-K4 methylation crosstalk.
We also identified several negative-regulators of H3K4me3 ( Figure 2C and 2D), including the one previous reported, Ctk1p (Xiao et al., 2007) and Jhd2p, a H3K4me3 demethylase (Liang et al., 2007). Interestingly, the lysine to alanine mutant H3K36A and loss of Set2p, a histone methyltransferase of H3K36 (see below), both resulted in elevated H3K4me3, suggesting a previously unknown cross talk between these two histone marks. Moreover, four components of the Rpd3 histone deacetylase complex (SIN3, EAF3, UME1, RCO1 and SAP30) and two HDA1 histone deacetylase complex members (HDA1 and HDA2) showed negative regulation on H3K4me3 ( Figure 2D).
These observed regulations are consistent with the previous finding that histone deacetylase inhibitors could enhance the level of H3K4me3 (Nightingale et al., 2007) and Gcn5 HAT complex enhances the level of H3K4me3 in transcribed coding sequences (Govind et al., 2007). In addition, several proteins were enriched to ribosomal subunit components (RPL22A, RPL42A, RPL8B, ASC1 and RPS29B) or those which are required for ribosomal subunit synthesis (NOP56, NOP8, MPP10, DIP2, DBP9 and YMR310C) ( Figure 2D). H3K4me3 is known to play a primary role in repression of ribosomal genes conjugating with H3S10p under multiple stress conditions (Weiner et al., 2012). Thus, it is possible that H3K4me3 and certain ribosomal proteins may have a feedback regulation, which is worth further investigation.
Overall, through a single CLICK array experiment, a general view of RMHs for H3K4me3 could be obtained ( Figure 2D). With their function annotations and previous studies, we can summarize the roles of known RHMs, and also explain the possible roles of the newly identified RHMs through their connections with the know RHMs.

Identification of the network of RHMs of H3K36me3 by the CLICK array
To further validate the CLICK array, another histone mark, H3K36me3, was examined ( Figure S3). Following a similar procedure to probe the array and analyze the data as described above, the fold-changes for all strains on the CLICK array were then determined. Overall, we identified 36 positive-RHMs and 54 negative-RHMs ( Figure   S3A) and some of them were verified by western blot (Figure S3B). Then we analyzed the interactions of these regulators as well as the known ones ( Figure S3C).
H3K36me3 is associated with transcriptional elongation and is enriched throughout gene coding regions, peaking at the 3' ends (Pokholok et al., 2005). As expected, besides Set2p, the histone lysine methyltransferase (KMT) for H3K36me3, several regulators related to RNA polymerase (RPB5 and RPB7) and transcription elongation factors (CTK1, CTK3, CDC73, SGV1 and TFG2) were identified ( Figure S3C). The result causing by deletion of YJL169W may due to its partial overlap with SET2. Many of these RHMs have been clarified for how to modulate H3K36me3 levels in previous studies (Wagner and Carpenter, 2012). For example, Set2p interacts with the phosphorylated C-terminal domain (CTD) of Rbp1p, the largest subunit of RNAPII, and this interaction is required for maintaining H3K36me levels (Kizer et al., 2005). Two subunits of Cterminal domain kinase I (CTDK-I), Ctk1p and Ctk3p, are required for H3K36 methylation as their roles in the phosphorylation of CTD of Rbp1p (Youdell et al., 2008).
In addition, the absence of Paf1p, Ctr9p and Cdc73p, the three components of Paf1C, results in a loss of H3K36me3 and the deletion of SGV1, which is required for the proper recruitment of Paf1C was observed to have the same effect (Chu et al., 2007).
We also found several ribosomal related proteins negatively associated with the level of H3K36me3 ( Figure S3C) but not H3K4me3. This may suggest that ribosomal proteins had different and specific roles in different histone marks. In fact, in mammal cells, the activation of KDM2A, demethylase for H3K36me1 and H3K36me2, could be inhibited when ribosome biogenesis is reduced under starvation (Tanaka et al., 2010). Our results indicated that in yeast there may be a similar regulation yet to be discovered.

Cab4p and Cab5p regulate the acetylation level of histone H4 on lysine 16
Given the capability of CLICK array to identify the regulators of well-studied modifications, we next apply the technology to discover modulators of H4K16ac, a functionally important but poorly studied histone mark.
Histone acetylation is known to be regulated by the opposing actions of HATs and HDACs. As the acetyl group has a negative charge, histone acetylation is expected to diminish the electrostatic affinity between histones and DNA, and thereby promote a more open chromatin structure that is more permissive to transcription (Shahbazian and Grunstein, 2007). H4K16ac is related to chromatin folding (Shogren-Knaak et al., 2006), heterochromatic silencing (Oppikofer et al., 2011) and cellular lifespan (Dang et al., 2009). In S. cerevisiae, H4K16 is acetylated primarily by the SAS complex, which consists of the catalytic subunit Sas2p and other two subunits, Sas4p and Sas5p (Suka et al., 2002). H4K16 can also be acetylated by the essential HAT Esa1p, which is also responsible for acetylating other H4 tail lysines (Bird et al., 2002). However, whether other factors are required for acetylation of H4K16 remains unknown.
A CLICK array was probed with specific antibodies against H4K16ac and H4 ( Figure   1B) and the data analyzed following a similar process as described above, resulting in the identification of 188 positive-regulators and 26 negative-regulators (Figures 3A).
Those with high fold-change were validation by western blotting (Figure 3B). To illustrate the relationships of these regulators, we analyzed their interactions and resulted some enriched sub-groups, including proteins related to aminoacyl-tRNA biosynthesis (YDR341C, GLN4, YHR020W, GRS1, YNL247W and GUS1), RNA polymerase subunits (RPC11, RPC25, RPC37, RPB5 and RPB8) and proteins related to mRNA slicing ( Figure. 3C). Interestingly, unlike H3K4me3 and H3K36me3, proteins related to ribosome showed positive regulations on H4K16ac ( Figure. 3C). This suggested that even though these three histone marks were all related to transcriptional activity, they were obvious by different regulated.
In addition, two essential genes, CAB4 and CAB5, which together catalyze the last two steps of the coenzyme A (CoA) biosynthetic pathway, were on the top list of We then focused on Cab4p and Cab5p. To clearly demonstrate the effects of Cab4p and Cab5p on the H4K16ac modification, cab4 and cab5 ts mutants were shifted from the permissive temperature (25 °C) to the non-permissive temperature (37°C) in YPD medium for 4 h. The samples were collected throughout this 4-h period (at 1, 2, 3, and 4 h) and subjected to western blotting using the H4K16ac antibody. Clear loss of H4K16 acetylation was observed for both cab4 and cab5 ts mutants after the shift to 37°C for as short as 1 h (Figures 4B, 4C). Almost complete loss of the H4K16ac modification was observed after the mutants were shifted to 37°C for 4 h (Figures 4B, 4C and 4D). To further confirm the roles of Cab4p and Cab5p in regulating H4K16ac, WT CAB4 and CAB5 were complemented in cab4 and cab5 ts mutants, respectively. The growth defects of the cab4 and cab5 ts mutants at the non-permissive temperature were restored in these strains (Figures S4A-D). At the same time, the loss of H4K16ac observed in the original ts strains was also restored (Figures 4E and 4F). These results strongly suggest that Cab4 and Cab5 are two key regulators of H4K16ac.

Cab4p and Cab5p play central roles in histone lysine acylation
Cab4p and Cab5p catalyze the last two steps of CoA biosynthesis (Olzhausen et al., 2009): Cab4p adds the AMP moiety to 4'-phosphopantetheine forming dephospho-CoA, and then Cab5p phosphorylates the 3'-OH of the ribose to yield CoA. CoA biosynthesis is an essential pathway and highly conserved (Leonardi et al., 2005). CoA is the precursor of acetyl-CoA and a variety of other acyl-CoA, such as crotonyl-CoA, propionyl-CoA and butyryl-CoA ( Figure 6A). We postulated that Cab4p and Cab5p may not only play roles in H4K16 acetylation, but also be critical in the acetylation of other histone sites and, further, other types of acylations. To test this hypothesis, cab4 and cab5 ts mutants were shifted to 37 °C and samples were obtained at different time 13 / 33 points. The cell lysates were analyzed by a variety of histone acylation antibodies, including H3K56ac, H4K8ac, H3K9bu, H3K14cr, H4K12cr, H4K12prop and H4K16bhb. In addition, two histone methylation specific antibodies H3K4me3 and H3K36me3 were also included. The amount of loaded proteins was determined by H3-or H4-specific antibodies. The results clearly showed that, similar to H4K16ac, there is a significant loss of all histone acylations tested but no change in the extent of histone methylations examined for both cab4 ( Figure 5A) and cab5 ts mutants (Figure 5B), while no significant changes were observed in the WT strain BY4741 (Figure 5C). To further confirm the roles of Cab4p and Cab5p in regulating histone acylation, ts mutant strains complemented with a plasmid carrying either WT CAB4 or CAB5, respectively, were also analyzed. As expected, the loss of the histone acylations were restored in strains at non-permissive temperature for both Cab4p ( Figure 5D) and Cab5p ( Figure 5E). We hypothesized that the loss of acylation may due to the depletion of CoA and acyl-CoAs in cab4 and cab5 ts mutants. In support of this argument, ultra-performance liquid chromatography-mass spectrometry (UPLC-MS) analysis of S. cerevisiae metabolites confirmed the significant loss of CoA, acetyl-CoA and butyryl-CoA in cab4 and cab5 ts mutants, and the loss could be at least partially restored by putting back the WT CAB4 and CAB5 (Figure 6B-D).
As cab4 and cab5 ts mutants affect the acylation of histone, we expected that cab1, cab2 and cab3 ts mutants would also affect histone acylation. Consistent with this, and supporting our results, previous work showed that reduced expression of Ppc1p, the homologue of Cab2p in Schizosaccharomyces pombe, caused diminished histone acetylation (Nakamura et al., 2012). However, we only detected a modest decrease in the level of histone acylation in the cab1 ts mutant but no in the cab3 ts mutant Cab5p not only result from their roles in CoA biosynthesis but also other unknown pathway.

Cab4p and Cab5p response differently to stress
To determine whether cab4 and cab5 ts mutants have other influence on cellular function, we tested cell growth under different stress conditions. We found that both cab4 and cab5 ts mutants were sensitive to 200 mM hydroxyurea (HU) not only at 37°C but also at 25°C (Figure S6A). The inhibition of 200 mM HU could be restored for the cab5 ts mutant, while not for the cab4 ts mutant at the non-permissive temperature ( Figure S6A). However, both cab4 and cab5 ts mutants were essentially insensitive to other DNA damage agents such as 10 μg/mL benomyl, 8 μg/mL camptothecin (CPT), 2 μg/mL nocodazole, 25 nM rapamycin and 3 mM H2O2 and to 100 J/m 2 ultraviolet (UV) radiation ( Figure. S6B). More interestingly, the cab5 ts mutant exhibited a growth defect on YPG and YPGE medium at the permissive temperature but cab4 ts mutant did not ( Figure. S6C) and only the growth of cab4 ts mutant could be restored on YPD containing 0.3% acetic acid at the non-permissive temperature (Figure. S6D). These results indicate that even though Cab4p and Cab5p were closely functioning in the same pathway, there are slight differences in metabolic regulations between them.

DISCUSSION
In this study, we constructed a CLICK array with the cell lysates from S. cerevisiae

Advantages and disadvantages of CLICK array
As compared with existing strategies, such as Global Proteomic Analysis of S. cerevisiae (GPS) (Krogan et al., 2003) and microwestern (Ciaccio et al., 2010), the CLICK array has several advantages. Firstly, it enables proteome-wide identification of regulators for any protein or PTM, i.e., histone marks. Since the array covers 96% of the nonessential genes and 26% of the essential genes, we are able to screen 82% of the proteome of S. cerevisiae in a single experiment. Secondly, the screening process is fast, only requiring 14-15 h. In comparison, the western blotting based strategy, GPS, takes at least several months to screen the regulators of a single histone mark. Thirdly, the CLICK array is highly reliable. The dual-color modification/histone labeling strategy, similar to that commonly used in DNA microarrays (Churchill, 2002), reduces many of the variations encountered in protein microarray experiments, such as batch-to-batch, array-to-array, or spot-to-spot variations. Simply calculating the Cy3/Cy5 ratio for each spot also greatly simplifies the data analysis. Fourthly, the CLICK array based strategy is physiological relevant. To make the CLICK array, the whole cell lysate is denatured and printed onto the array directly. Theoretically, most, if not all, of the proteins from a strain will be immobilized on a single spot, providing a "snapshot" of the particular physiological proteomic status. We thus expect that the level and the pattern of the histone marks are well-retained on the CLICK array. Likewise, therefore, any regulators identified are anticipated to be physiologically important. Lastly, the CLICK array is a high-throughput strategy. Usually, it consumes 0.3-0.5 nL of sample per spot. Thus, the cell lysate prepared from one batch of 1 mL culture is enough to print 1,000 to 10,000 CLICK arrays, which lowers the cost for a single microarray to negligible. The printed arrays can be stored and later probed individually or in batch whenever necessary.
To make the CLICK array strategy generally applicable, several limitations have to be 16 / 33 overcome. Firstly, it is not easy to obtain well-qualified antibodies, for example, a histone mark-specific antibody suitable for CLICK array assay. Even though there are more than 1,000 commercially available antibodies for various histone marks (Rothbart et al., 2015), recent studies have revealed many unpredicted and alarming observations regarding the properties of the antibodies, including non-specificity, strong influence by adjacent PTMs, and inability to distinguish the modification state on a particular residue (e.g., mono-, di-, or tri-methyl lysine) (Bock et al., 2011;Egelhofer et al., 2011). However, this is indeed now a well-recognized problem, with many focused efforts working on its resolution (Hattori et al., 2013;Rothbart et al., 2015). We anticipate that more histone mark-specific antibodies of CLICK array-quality will soon emerge. Secondly, the lysates on the current CLICK array represent only a single culture condition. Since histone marks can change under different culture conditions, some modifications may be absent or under represented on the current CLICK array. This possibility could be easily overcome by preparing several sets of cell lysates at different culture conditions, such as investigating mid-log vs stationary phase, adding sodium butyrate to improve the acetylation level (Davie, 2003), or adding nocodazole to improve the phosphorylation level (Baker et al., 2010). All of these lysates could then be spotted onto a single array or a set of arrays, enabling a more profound screening of regulators for histone marks. Lastly, many of the regulators identified by the CLICK array could be indirect regulators, that is neither "writer" nor "eraser". It may be difficult to further dissect their molecular mechanisms. To overcome this issue, bioinformatics, proteomics or other techniques, e.g. protein microarray could be applied to provide homolog or interacting proteins.

Toward a comprehensive regulatory and functional network for histone marks
The data set presented here provides a comprehensive view of RHMs for several histone marks. For studying their functional roles in biological progress such as transcription, specific histone residue mutant could help us to know which genes they could influence. With powerful genomic technologies such as DNA microarray and 17 / 33 RNA sequencing, it is easy to globally reveal the downstream consequence of an alteration on a given gene or protein. Thus, combining our CLICK array results and gene expression dataset, we could, for the first time, obtain a network which contains both the upstream regulators determining the abundance or PTM of certain protein, i.e., histone mark and the downstream global consequence in response to the corresponding specific interruption, i.e., histone residue mutant, in another word, a "complete" regulation picture of a given protein or PTM (Figure 7, S8 and S9). For example, in H3K4A mutant strain (GSE29059) (Jung et al., 2015), the down expression genes were clustered to ribosomal biogenesis while the up expression genes were majorly located in mitochondria and related to energy derivation. However, the expression levels of the upstream regulators of H3K4me3 that we identified have not been obviously altered, suggesting that the transcription of these upstream regulators are not affected by H3K4A mutant or be masked by backup mechanisms.

H4K16ac
In this study, we have built three global regulating networks with both the upstream and downstream regulators for H3K4me3, H3K36me3 and H4K16ac (Figure 7, S8 and   S9). We found there are several regulators shared among these histone marks ( Figure   S7).
For H3K4me3 and H3K36me3, by CLICK array we found they shared 4 positiveregulators, 6 negative-regulators, and 4 regulators demonstrated the very reverse effect for these two histone marks (Figure S7A). Since some of the known regulators of H3K4me3 and H3K36me3 were missing from the CLICK array, these known regulators were then put together with the regulators that we identified on the CLICK array and subjected for cross talk analysis ( Figure S7B). Indeed, Paf1C and BUR kinase complex has been reported to be required for H3K4me3 and H3K36me3 (Chu et al., 2007;Krogan et al., 2003;Laribee et al., 2005). It is worth noting that 4 positiveregulators for H3K36me3 cause negative effect on H3K4me3. Previous study 18 / 33 demonstrated that the deletion of CTK1 elevated H3K4me3 level and this is related neither to COMPASS and Paf1C nor to Rad6/Bre1-mediated histone H2B monoubiquitination (Wood et al., 2007). The interaction between Set2p and Pol II is due to phosphorylation of Pol II CTD by the CTK kinase (Xiao et al., 2003). Our results indicates that the regulation of H3K36me3 by CTK1 results in its negative regulation role on H3K4me3. In addition, proteins associated with histone deacetylation, Sin3p and Eaf3p, were found to affect H3K4me3 levels ( Figure 2C). Several studies have showed that H3K36me could be recognized by Eaf3p and recruits the Rpd3S deacetylase complex, and leads to deacetylation of H3 (Carrozza et al., 2005;Joshi and Struhl, 2005;Keogh et al., 2005). H3K4 methylation also facilitates histone acetylation (Wang et al., 2009) suggesting a positive feedback loop between histone acetylation and H3K4 trimethylation. Both H3K4 methylation and H3K36 methylation are considered to be associated with transcriptionally active genes and H3K4me3 peaks at the beginning of the transcribed portions of genes while H3K36me3 is abundant throughout the coding region, peaking near the 3' ends (Pokholok et al., 2005).
Recruitment of deacetylase by Set2p mediates H3K36 methylation, prevents spurious transcription within coding regions (Lee and Shilatifard, 2007), and may eventually lead to the loss of H3K4me3 to inhibit inappropriate initiation within coding regions.
Altogether, our results and previous studies suggest a strong crosstalk between H3K4me3 and H3K36me3 through their different roles in histone acetylation regulation.
For H3K4me3 and H4K16ac, we found they shared 3 positive-regulators, and 10 regulators demonstrated the very reverse effect for these two histone marks ( Figure   S7C and D). In fission yeast, loss of Leo1p, a component of Paf1C, would lead to reduced level of H4K16ac at heterochromatin boundary regions (Verrier et al., 2015).
The recruitment of Paf1C is depend on the phosphorylation of the C-terminal repeats (CTRs) of Spt5 by Sgv1p (Liu et al., 2009). As Cdc73p is a component of Paf1C, we speculate that the decrease of H4K16ac level caused by deletion of CDC73 and SGV1 may due to the similar mechanism. Moreover, we also found in H4K16A mutant strain, 19 / 33 the level of H3K4me3 had a modest increase, as well as in H3K56A and H4K5A mutant strains, but not for H3K9A, H4K8A and H4K12A, even though these residues could be actylated. Even though H3K4me3 can trigger acetylation and deacetylation of histone H3 and H4 in certain contexts (Latham and Dent, 2007) and several positive regulatory residues on histones are required for the implementation of normal levels of H3K4me3 (Nakanishi et al., 2008), it is little known about the negative regulatory residues for H3K4me3. The CLICK array provides a method for systematically profiling and comparing of regulatory residues on histones for certain histone mark since most of the histone lysine to alanine mutants are avaliable (Dai et al., 2008).
In addition, we found that the N-terminal acetyltransferase subunit Ard1p also affects H3K4me3 (Figures 2B). Ard1p is required for maintaining levels of H3K79me3 and H2B monoubiquitination (Takahashi et al., 2011). As histone H3K4 and H3K79 methylation is dependent on H2BK123 monoubiquitination (Nakanishi et al., 2009), Ard1p may similarly affect the trimethylation of H3K4 through regulation of H2BK123 monoubiquitination. The relationship between Ard1p and H3K4me3 needs to be further characterized.

The roles of Cab4p and Cab5p in histone acylation
In conventional analysis, identification of new regulator is usually based on proteinprotein interactions that may miss indirect ones (Ji et al., 2015). In this study, we found two indirect regulators, Cab4p and Cab5p, have a significant impact on H4K16ac (Figure 4). They are responsible for catalyzing the last two steps of CoA biosynthesis (Olzhausen et al., 2009). CoA is an essential cofactor for a large number of enzymes involved in the transfer of acyl groups and the metabolism of carboxylic acids and lipids in all organisms (Leonardi et al., 2005). Pantothenate is the donor for CoA biosynthesis.
Cab4p adds the AMP moiety to 4'-phosphopantetheine forming dephospho-CoA, and then Cab5p phosphorylates the 3'-OH of the ribose to yield CoA. Acetyl-CoA, as the predominant CoA thioester, has been reported to have an influence on histone acetylation, and acetyl-CoA synthetase is also required for histone acetylation (Cai et 20 / 33 al., 2011;Takahashi et al., 2006). As the depletion of CoA will have a global effect on the concentration of the acetyl-CoA, we hypothesized that the same effect on other histone acetylation sites, and even other kind of acyl modifications, might be observed.
Consistent with this, we found that Cab4p and Cab5p affects all of the histone acetylation and acylation reactions that we have investigated, namely H3K56ac and H4K8ac, as well as butyrylation, propionylation, crotonylation, and βhydroxybutyrylation. As expected, we found no effect on histone methylation in the cab4 and cab5 ts mutants (Figure 5), consistent with the idea that these mutants have their effect via a deficit of CoA. The UPLC-MS based metabolic analysis further supports our hypothesis that the reduced CoA level, resulting from the incomplete biosynthetic pathway, results in the depletion of acyl-CoA, which are the donors of histone acyl modifications. Therefore, the biosynthesis of CoA plays a critical role in histone acylation.
We note that the CoA biosynthesis pathway is conserved from bacteria to mammalian (Aghajanian and Worrall, 2002;Martinez et al., 2014). In fact, the null mutants of enzymes in this pathway in S. cerevisiae can be complemented by their bacterial counterparts (Olzhausen et al., 2009). CoA and acetyl-CoA participate in many biology processes and virtually every enzyme involved in glycolysis, fatty acid metabolism and the TCA and urea cycles is acetylated (Theodoulou et al., 2014).
However, we did not observe obvious differences in acetylation levels of non-histone proteins in cab4 and cab5 ts mutants as compared to that of WT strains ( Figure S10A).
The same result was observed when knocking out coaE in E. coli, which does not have histone proteins (Figure S10B). These results indicate that the depletion of CoA caused by the cab4 and cab5 ts mutations only significantly affects acylation of histones but not other non-histone proteins.

Perspectives of the CLICK array
The CLICK array can be easily extended for wider applications. In this study, we investigated thousands of gene knockouts or TS mutants. Thousands of other 21 / 33 conditions or cell states could be similarly investigated, such as chemical/drug treatment, culture conditions, or overexpression of single genes. Further, the application of the CLICK array is not limited to the discovery of regulators for histone marks, and it could be easily adopted for studying upstream regulators of other proteins or biological molecules, as long as high-affinity reagent is available. Though S. cerevisiae is a highly effective model for the CLICK array, this array is also readily applicable to other species, including human. To prepare a collection of cell lines of kilo-conditions, CRISPR/Cas9 (Shalem et al., 2014), or RNAi (Echeverri and Perrimon, 2006) are two practical options. In addition, the histone marks and their specific antibodies are much more extensive in human than in S. cerevisiae. We anticipate that once the human CLICK array is available, it will be a very powerful tool for wide range of studies.
Taken together, to accelerate the discovery of regulators and the functional study for histone marks, taking advantage of the high-density array format and the specific antibodies, herein, we established the CLICK array strategy. Using S. cerevisiae as a model, we built regulatory networks for three histone marks and re-discover the majority of known regulators for H3K4me3 and H3K36me3. In a test for new factors, we showed that Cab4p and Cab5p, the last two enzymes in the CoA biosynthesis pathway play central and surprisingly specific roles in histone acylation through controlling the level of acyl-CoA. The CLICK array strategy is generally applicable for the identification of regulators of other histone marks and other proteins and it is also readily extensible for other species and biomolecules.

Antibodies
A complete list of antibodies used in our experiments were purchased from various sources (see Supplementary Table 1) or information on the antibodies. All antibodies were used at a dilution of 1:2,000 for western blotting, 1:200 for dot blotting and 1:1000 for microarray.

Cell culture and lysate preparation
For gene deletion mutants, strains were inoculated from -80 °C stocks onto agar plates containing YPD + 200 μg/mL GENETICIN (Thermofisher), allowed to grow 48 h at 30 °C.
Then inoculate yeast cells from agar plates to a 96-well 2 ml box in which every well contained 500 μL YPD liquid media and a 2 mm diameter glass ball, which facilitated 23 / 33 the uniform growth. After 24 h of growth at 30 °C, transfer to 6 ml fresh YPD. Grown at 30 °C with vigorous shaking about 16 h, the culture should reach O.D.600 1.0-1.2. The cells were harvested by spinning at 3000 g for 5 min, and the cell pellets were washed with 500 μL water one time. The washed semi-dry culture was immediately stored in -80 °C freezer. For ts mutants, cells were grown in URA-/dextrose liquid media and reached to O.D.600 0.5-0.6 at 25 °C, then shifted to the nonpermissive temperature for 2 h. Frozen cells were thawed at room temperature, resuspended in 1 mL 0.1 M NaOH and incubated at room temperature for 10 min. After centrifugation, 60 μL SDS sample buffer (0.6 M Tris pH 6.8, 0.5% SDS, 20% glycerol, 2% β-mercaptoethanol) was added and the samples were vortexed briefly before heating at 95 °C for 10 min. The supernatants of yeast cell lysate were used for microarray and western blotting and stored at -20 °C.

Cell lysate microarray fabrication
We transferred lysates to wells in 384-well polypropylene plates (10 µl/well). We used a contact-printing robotic SmartArrayer 48 microarrayer (CapitalBio) fitted with solid spotting pins to spot lysates onto FAST slides (Schleicher & Schuell BioSciences). Slides coated with a single nitrocellulose pad and each cell lysate spotted twice. The resulting microarrays were stored at -20 °C prior to use.

Probing antibodies on the cell lysate microarray
CLICK array were blocked for 1 hr at room temperature with shaking in blocking buffer (3% BSA in Tris-buffered saline solution containing 0.1% Tween 20 detergent [pH 7.4]).
After blocking, arrays were probed with 3 ml site-specific histone mark antibody (PTM Lab) and at the same time, H3 or H4 antibody from different species would be added as its loading control. After incubating overnight at 4 °C with shaking, arrays were washed three times with shaking in 1xTBST and then probed with 3 ml Cy5-donkeyanti-rabbit antibody and Cy3-donkey-anti-mouse antibody (1 μg/ml in blocking buffer) for 1 h at room temperature. After washing three times in TBST, arrays were dried in a 24 / 33 SlideWasher (CapitalBio) and then scanned with a GenePix 4200A microarray scanner (Molecular Devices). Data were analyzed with GenePix Pro 6.0 (Molecular Devices).

Extraction of CoA and acyl-CoA
Saccharomyces cerevisiae cells were grown in the SC-Met medium to O.D.600 1.4-1.6, and 70 ml of the culture was used for the metabolite extraction. After washed with ddH2O, the harvested cells were suspended in 500 μL of cold 80% (v/v) methanol (-40 °C). After ultrasonic crashing for 5 min, the cells were frozen at -80 °C for 30 min and then put on ice to thaw the cell suspension. Freeze-thaw cycles were repeated five times in order to extract the cells. The methanolic extracts were separated from the cell debris by centrifugation at 16,000 rpm at 4 °C for 15 min. The supernatant were evaporated to dryness and dissolved in 50 μL of 40% methanol containing 1 μg/mL Acetyl-1,2-13 C2-CoA as internal standard for quantification.

Array analysis
We used GenePix Pro 6.0 to determine the median pixel intensities for individual 25 / 33 features and background pixels in both Cy3 and Cy5 channels. All the data was analyzed by limma package in R. Two chanel color signals were normalized using "loess" algorithm (Smyth and Speed, 2003). We normalized the backgroundsubtracted Cy5 intensities (site-specific histone mark antibody) to the level of histone (Cy3) at each feature by taking the ratio of Cy5 to Cy3. The value was used for all subsequent analysis. We calculated fold change with respect to BY4742 for each positive spots. P values was calculated using z-score test with non-parameter statistical method. Briefly, to avoid the fold-change bias, the median value and the median absolute deviation (MAD) value were taken as robust mean value and robust standard deviation, respectively. The all fold-change values were transformed to z-scores by the following formula: z-score = (fold-change -median(fold-change))/MAD(fold-change). P values were then calculated based on z-scores by applying standard normal distribution with two-tailed alternative hypothesis.

Data Resources
Raw data files of the CLICK arrays profiling with H3K4me3, H3K36me3 and H4K16ac specific antibodies have been deposited on http://www.protein--microarray.com/ as accession number: PMDE230.   respectively. This plot is colored such that those points having a fold-change less than 0.5 (log2 < 0.5) or above 0.5 and p value less than 0.01 simultaneously, are shown in green and red, respectively.