Small proteins modulate ion channel-like ACD6 to regulate immunity in Arabidopsis thaliana

ACCELERATED CELL DEATH 6 (ACD6) mediates a trade-off between growth and defense in Arabidopsis thaliana. However, the precise biochemical mechanism by which ACD6 and related proteins in plants act remains enigmatic. Here, we identified two loci, MODULATOR OF HYPERACTIVE ACD6 1 (MHA1) and its paralog MHA1-LIKE (MHA1L), that code for ∼7 kDa proteins that differentially interact with specific ACD6 variants. MHA1L enhances accumulation of an ACD6 complex, thereby increasing activity of the ACD6 standard allele for regulating plant growth and defenses. ACD6 is a multipass transmembrane protein with intracellular ankyrin repeats that are structurally similar to those found in mammalian ion channels. Several lines of evidence link increased ACD6 activity to enhanced calcium influx, likely mediated by ACD6 itself and with MHA1L as a direct regulator of ACD6.


INTRODUCTION
While plants need to defend themselves against pathogens, an inherent danger is inappropriate activation of immune responses, which can cause collateral damage to the plant itself. Particularly effective immune alleles enable a plant to respond rapidly to pathogen attack, but such variants might also be potentially deleterious.
The study of natural variation in the immune system of Arabidopsis thaliana has identified ACCELERATED CELL DEATH 6 (ACD6), a positive regulator of cell death and defense responses, as a nexus for a trade-off between growth and disease resistance in wild populations [1][2][3] . The natural ACD6-Est-1 allele can protect plants against a wide range of unrelated pathogens, but at the same time often exacts a substantial growth penalty in form of reduced stature and cell death in leaves 1 .
The hyperimmunity conditioned by high levels of ACD6 activity is dependent on salicylic acid (SA), with inactivation of individual SA signaling components partially suppressing the effects of ACD6 hyperactivity 4,5 . SA signaling is required for several aspects of plant immunity, including pattern-and effector-triggered immunity (PTI and ETI) as well as systemic acquired resistance (SAR) [6][7][8][9] . Both SA biosynthesis and many aspects of SA up-and downstream signaling are well understood, with feedback and feedforward regulation being important mechanisms. ACD6 stimulates SA accumulation, and SA in turn enhances ACD6 mRNA expression and affects the subcellular localization of ACD6 protein 10,11 . A working model for ACD6 is that it is a redundant regulator of SA signaling through a positive SA-dependent feedback loop [10][11][12] . An ACD6 homolog, BDA1, also functions as a positive regulator of immunity in A. thaliana 13 , as do maize and wheat ACD6 homologs, which have recently been implicated in smut (ZmACD6) and leaf-rust (Lr14a) resistance, respectively 14,15 .
ACD6 encodes a multipass transmembrane protein with nine intracellular ankyrin repeats 10,11 . ACD6 protein has been found in association with multiple plasma membrane-localized pattern recognition receptors (PRRs) and the PRR co-receptor BRI1-ASSOCIATED RECEPTOR KINASE 1 (BAK1) as well as other receptor like kinases, supporting a direct role for ACD6 in plant immunity 11,12,16 . In response to SA signaling, ACD6 oligomers accumulate at the plasma membrane, indicating that plasma membrane localisation of ACD6 is important for its function 11,12 . ACD6 interacts genetically also with the TIR-NLR receptor gene SUPPRESSOR OF NPR1-1, CONSTITUTIVE 1 (SNC1), with the SNC1 weak allele attenuating the atutoimmune phenotype of ACD6-Est-1 in some wild strains of Arabidopsis 17 . The ACD6 homolog Lr14a in wheat could trigger Lanthanum(III) chloride inhibited water-soaking phenotype in N.benthamiana leaf 15 , suggesting role of ACD6 homologs in regulating calcium fluxes. However, the precise biochemical mechanism by which ACD6 and related proteins act remains enigmatic.
There has recently been considerable progress about a series of plant immune proteins that regulate calcium influx during pathogen responses 18 . These include classical ion channels, such as glutamate receptorlike proteins (GLRs) 19 , cyclic nucleotide-gated ion channels (CNGCs) 20 , reduced hyperosmolality-induced [Ca 2+ ]cyt increase channels (OSCAs) 21 , as well as non-canonical NLR-formed ion channels 22,23 . Here, starting from the analysis of natural modifiers of ACD6 activity, we identify a new gene family encoding small proteins, MODIFIER OF HYPERACTIVE ACD6 1 (MHA1) and its paralog MHA1-LIKE (MHA1L), which affect ACD6 activity in a complex manner. Sequence and structure similarity of ankyrin repeats in ACD6 with those of transient receptor protein (TRP) ion channels from animals and fungi spurred us to investigate whether MHA1 and/or MHA1L effects might be mediated by ion-dependent ACD6 signaling. We find that increased ACD6 activity enhances calcium signaling, most likely because ACD6 itself is an ion channel. Since MHA1 and MHA1L both bind to the ankyrin repeats of ACD6, we propose that MHA1 and MHA1L are ACD6 ligands that regulate ACD6-dependent ion channel activity and signaling.

Identification of MODIFIER OF HYPERACTIVE ACD6 (MHA) loci
Many, but not all A. thaliana accessions carrying the ACD6-Est-1 allele express visible signs of autoimmunity in the absence of pathogen challenge 1 . Using genome-wide association study (GWAS) of accessions with the ACD6-Est-1 allele, we identified two unlinked regions, MODIFIER OF HYPERACTIVE ACD6 1 (MHA1) (Chr1: 22,935,037, p=10 -12 ), and MHA2 (Chr4: 11,019,243, p=10 -8 ) ( Figure 1A), which together explained over 60% of variation in macroscopic cell death among ACD6-Est-1 carriers. To confirm the GWAS results, we made use of the accession Ty-0, which has cell-death suppressing alleles at both MHA1 and MHA2 ( Figures 1B   and S1A). There were no nonsynonymous differences in ACD6 between Ty-0 and Est-1, and ACD6 was well expressed ( Figure S1B), but Ty-0 had much less SA than Est-1, and the SA-responsive marker gene PATHOGENESIS-RELATED 1 (PR1) was also expressed at much lower levels ( Figures S1B and S1C). In support of phenotypic differences mapping outside of ACD6, ACD6-Ty-0 and ACD6-Est-1 similarly triggered necrotic lesions and reduced biomass when introduced as transgenes into the Col-0 reference background ( Figures S1D and S1E). Together, these results demonstrate that Ty-0 harbors extragenic suppressors that modulate the activity of ACD6-Est-1.
The top GWAS hit in the MHA1 region residues in AT1G62045, which encodes a small ~7 kDa protein of 67 amino acids. To examine whether AT1G62045 regulates ACD6-Est-1 activity, we overexpressed in the Ty-0 background the MHA1 alleles of Ty-0 and Est-1 ( Figure 1C), which differ in two adjacent codons, with one of these causing a non-conservative change from asparagine to serine due to the top GWAS SNP ( Figure   1D). Overexpression of MHA1-Est-1 but not MHA1-Ty-0 caused strong leaf necrosis and reduced size in 13 out of 15 T1 individuals, supporting that AT1G62045 is MHA1 ( Figure 1C). MHA1 homologs are found in many plant species (Figure S1F), and alignment of the predicted protein sequences told us that other Brassicaceae typically encode the asparagine corresponding to the Est-1 variant, which is representative of the major A. thaliana allele ( Figure 1D). The Ty-0 allele of MHA1 carried a nonsynonymous substitution that affected a conserved asparagine ( Figure 1D), suggesting that this allele might have reduced or no activity, in agreement with the absence of effects in the overexpresssion experiment. Knockout of MHA1, however, led to cell death and increased PR1 expression in Ty-0, indicating that MHA1-Ty-0 was functional ( Figure 1E). This was surprising, because the MHA1-Est-1 allele seemed also to be functional, since overexpression of MHA1-Est-1 in Ty-0 also caused cell death and reduced growth ( Figure 1C). Conversely, introduction of a genomic copy of MHA1-Ty-0 into Est-1 mha1 mutants suppressed cell death and increased biomass, in agreement with this allele interfering with the action of ACD6-Est-1 ( Figure S1G). Together, these data pointed to MHA1-Ty-0 as a gain-of-function allele that acts as a negative regulator of ACD6-Est-1 activity in cell death promotion. The standard allele, MHA1-Est-1, apparently masks the action of MHA1-Ty-0, but it is not required for ACD6-Est-1 activity.

Genetic characterization of the MHA1 paralog MHA1L
In the A. thaliana reference genome, MHA1 and the adjacent gene AT1G62050 are paralogous to the 3' and 5' portions of another gene on chromosome 1, AT1G11740. This gene model appears to be mis-annotated, as public RNA-seq data and RT-PCR analyses indicated that two genes are transcribed from AT1G11740 (Figures S1H-S1K). We thus designated the MHA1 paralog MHA1-LIKE (MHA1L). The gene duplication that gave rise to MHA1 and MHA1L appears to have occurred at the base of the Brassicaceae as part of a whole-genome duplication 24 , and independent duplications are found in other lineages (Figure S1F). A mha1 mha1L double mutant in Est-1 resembles the wild type plants in ADC6-mediated cell death, suggesting that neither MHA1 nor MHA1L is required for ACD6-Est-1 activity ( Figure S1L).
As overexpression of MHA1-Est-1 constitutively activates immune responses, we further characterize the roles of MHA1 and MHAL1 in plant immunity. An expression atlas 25 revealed that MHA1 expression is elevated when plants are infected with the bacterium Pseudomonas syringae pv. tomato (Pst) DC3000 expressing avrRPM1 or with the fungus Botrytis cinerea. MHA1L expression was also induced by Pst DC3000 infection (Figures S2A). To gain further insights into MHA1 family function in immunity, we generated mha1, mha1l, and mha1 mha1l double mutants in the Col-0 reference background, which has the standard ACD6 allele (Figures S2B). Reactive oxygen species (ROS) burst and MAPK activation after flg22 treatment were weaker in mha1l and acd6-2 null mutants than in wild type (Figures 2A, 2B, and S2C), suggesting that MHA1L, like ACD6, is involved in PTI. In agreement, mha1l, mha1 mha1l, and acd6-2 mutants were similarly hypersusceptible to Pst DC3000 and Pst DC3000 hrcCstrain compared to Col-0 plants ( Figures 2C and 2D).
However, neither mha1l nor mha1 mha1l had obvious defects in ETI ( Figures S2D and S2E). There was a small change in ROS burst and MAPK activation, but not in Pst DC3000 growth in mha1 single mutants (Figures 2A-2D and S2C-S2E), suggesting that MHA1 does not play a major role in defense in the presence of the ACD6 reference allele in Col-0.
We also compared the effects of increased MHA1 and MHA1L activity in the reference accession Col-0. Neither MHA1-Est-1 nor MHA1-Ty-0 overexpression had obvious effects in Col-0, although MHA1-Est-1 had slightly higher PR1 expression ( Figure S2F). In contrast, overexpression of MHA1L, which does not show allelic variation in amino acid sequence among Col-0, Est-1, and Ty-0, caused strong cell death and dwarfing at 16°C in 24 out of 30 T1 plants ( Figure S2G). The effects at 23°C were milder ( Figure S2G), which can be at least partially attributed to reduced accumulation of the protein at the higher temperature, as demonstrated with plants overexpressing an GFP-MHA1L fusion ( Figure S2H). The temperature sensitivity of the MHA1L overexpressors suggested that MHA1L can enhance ACD6 activity, since it resembled the behavior of acd6-1 gain-of-function mutants, which also show enhanced cell death at 16°C compared to the standard temperature of 23°C (Figures 2E and S2E) 2 . Similarly, MHA1L overexpressors in mha1l showed enhanced ROS production, MAPK activation and resistance to Pst DC3000 and Pst DC3000 hrcCbacteria ( Figures 2E-2I), in support of a role of MHA1L in PTI. Genetic analysis confirmed that MHA1L overexpression defects were dependent on SA, as the cell death phenotype was either partially or completely suppressed by mutations in key SA signaling components such as NPR1 and helper NLRs of the ADR1 and NRG1 families (Figures S3A-S3C).

Identification of ACD6 as a potent suppressor of MHA1L overexpression
To illuminate the role of MHA1L in immunity, we generated a mutagenized population of MHA1L overexpression in an unbiased manner, we mutagenized MHA1L overexpressors in the Col-0 background with ethyl methanesulfonate and screened for suppressors ( Figure 3A). In total, revealed that 37 out of the 72 suppressor lines carry mutations in ACD6 ( Figure 3B). Four mutations in ACD6 were reported to be revertants of the acd6-1 gain-of-function allele 26 , consistent with them being loss-of-function alleles. Thirteen mutations affected residues conserved between ACD6 and its wheat homolog Lr14a, including L300F, which corresponds to L362F in Lr14a, which is a loss-of-function mutation in wheat 15 (Figure 3B,). In addition, cosegregation analysis showed that the ACD6 locus co-segregated with suppression of MHA1L-associated cell death for one of somha1l lines ( Figure S3D). Final validation for ACD6 being essential for the effects of  p-values from Tukey's HSD test. Scale bars 1 cm. See also Figure S3, and Table S1.
Together, the experiments with MHA1L overexpressors and acd6 loss-and gain-of-function alleles in the Col-0 background indicated that MHA1L function depends on ACD6 activity. However, not only was the ACD6-Est-1 phenotype not enhanced by MHA1L overexpression, but the severe MHA1L overexpression phenotype at 16°C was suppressed in Est-1 and in a NIL with the ACD6-Est-1 allele (Figures S3E and S3F), indicating genetic interaction of MHA1L with the standard ACD6-Col-0, but not the ACD6-Est-1 allele ( Figure   S2G). Note that the phenotype of ACD6-Est-1, in contrast to that of acd6-1, is not enhanced by lower temperature 1,2 , which distinguishes ACD6-Est-1 from many other autoimmunity mutants 27,28 . The genetic interactions and temperature effects between MHA1/MHA1L and ACD6 alleles are summarized in Figures   S3G.

Characterization of interaction between ACD6 and MHA1L
The genetic reliance of MHA1L on ACD6-Col-0 in the Col-0 accession prompted us to test whether their encoding proteins interact. MHA1L interacts with both full length and the cytoplasmic ankyrin repeat fragment of ACD6 in split-luciferase complementation assays in Nicotiana benthamiana and co-immunoprecipitation (co-IP) in A. thaliana protoplasts ( Figures 4A-4C). Direct interaction between MHA1L and ACD6-Col-0 was further observed in the in vitro GST or MBP pull-down assays ( Figures 4D and 4E). In co-IP and GST pulldown assays ( Figures 4C and 4D), MHA1L tended to produce a stronger signal than MHA1, but more quantitative methods would be required to determine whether these apparent differences are significant.
Finally, we examined the subcellular localisation of MHA1 and MHA1L fused with GFP tag in Col-0. In both leaf and root cells, GFP signal was detected both in the cytoplasm and at the plasma membrane ( Figures   S4A-S4D). Application of endocytosis inhibitor brefeldin A (BFA) enables GFP-MHA1 to co-localize endocytic tracer FM4-64 in the BFA bodies, suggesting that GFP-MHA1 recycles between the plasma membrane and endosomal compartments, similar to many other plant plasma-membrane proteins 29 ( Figure   S4B). Subcellular fractionation indicated that both GFP-MHA1 and GFP-MHA1L proteins were enriched in microsomes ( Figures S4E and S4F), similar to ACD6-1 protein 11 ( Figure S4G). The similar subcellular localization of MHA1, MHA1L and ACD6 was in agreement with direct interaction of the proteins.
We developed the following hypothesis for the MHA1/MHA1L interaction with ACD6: First, activity of the standard form of ACD6 (as found in Col-0) is enhanced by MHA1L; such an enhanced activity could provide additional feedforward regulation of ACD6 in response to pathogen challenge, as both enhanced ACD6 activity and infection of plants with a bacterial pathogen increased MHA1L mRNA accumulation ( Figure   S3A). Second, the amino acid substitutions in the transmembrane portion that are causal for increased activity of ACD6-Est-1 1 alter the conformation of ACD6-Est-1, such that it no longer requires binding of MHA1L to its ankyrin repeats for increased activity. ACD6-Est-1 can, however, be bound by MHA1-Ty-0, which in turn interferes with ACD6-Est-1 activity. Since MHA1L overexpression had more drastic phenotypic effects at 16°C than at 23°C (Figures 3B and S3B-S3F), we performed co-IP experiments at both temperatures. We found that higher temperature not only reduced ACD6 accumulation, but also ACD6 interaction with MHA1L ( Figure 4G).
We further examined ACD6 accumulation in A. thaliana plants using Blue Native polyacrylamide gel electrophoresis (BN-PAGE), which has been previously used to show that ACD6 exists in complexes of around 700-800 kDa 11,16 . Overexpression of MHA1L increased the total levels of ACD6 and that of large ACD6 complexes, which was also enhanced by lower temperature (Figure 4H), similar to what has been shown for the mutation in acd6-1, which, like MHA1L overexpression, enhances ACD6 activity 10,11 . Finally, analysis of complexes with 2D SDS-PAGE confirmed that MHA1L interacts with ACD6 and promotes ACD6 complex formation, with MHA1L and ACD6 being present in high-molecular-mass complexes of similar sizes ( Figure 4I).

MHA1L enhances ACD6-stimulated ion channel activity
Genetic and biochemical experiments had indicated that MHA1L strongly modulates ACD6 activity. To glean more clues about ACD6's function, we performed HHpred profile Hidden Markov Model searches 30 . We found multiple excellent hits to transient receptor potential (TRP) channels from animals and fungi, which regulate ion flux in response to stimuli ranging from heat to natural products and proinflammatory agents 31,32 ( Figure S5A). Alignments revealed extensive similarity between the ankyrin repeats of ACD6 and TRP proteins ( Figure S5B), but not between their transmembrane domains 33 . Out of 12 loss-of-function mutations mapping to the ACD6 ankyrin repeats, eight of the affected residues were identical in alignments to at least two out of the three top HHPred hits, fly NOMPC, human TRPA1, and rat TRPV6 ( Figure 3B).
The A. thaliana genome encodes over 100 ankyrin repeat proteins, with many of them having transmembrane domains 34 . We could reliably identify five transmembrane domains in ACD6 35 . In contrast, TRP channels have six transmembrane helices, with helices 5 and 6 forming the ion pore in a functional TRP multimeric ion channel 31,32 . Homology modeling did not suggest that the structure of the ACD6 transmembrane domains is consistent with those of TRP channels ( Figures S5C and S5D).
Although the similarities between TRP and ACD6 were restricted to the ankyrin repeats, ACD6 has multiple transmembrane domains that might form a pore, and we therefore investigated potential channel activity of ACD6 using African clawed frog (Xenopus laevis) oocytes. Since calcium has been implicated in different steps of immune signaling 36-38 , we were particularly interested in potential ACD6 activity as a calcium channel or regulator of calcium channels. Unfortunately, direct observation of calcium influx is hampered in Xenopus oocytes by the presence of endogenous calcium-activated chloride channels, which are stimulated by the accumulation of intracellular calcium and typically mask small ionic currents from calcium influx. However, these endogenous channels can be used as quantitative readout for channel-mediated calcium influx by foreign functional calcium channels 20,39 .
When we raised the external calcium concentration in our experiments with oocytes expressing standard ACD6 from Col-0 or the gain-of-function ACD6-1 variant, we observed ionic currents that resembled calcium-activated chloride currents in their current-voltage relationship; these currents were lost upon removal of external chloride ( Figures 5A and 5B). The calcium-activated chloride currents were much lower upon expression of the ACD6 L557F variant, which has a substitution in the transmembrane domain ( Figure 5C). Consistent with the role of MHA1L as an ACD6 activator in plants, MHA1L could further enhance the currents in ACD6-expressing oocytes ( Figure 5D). This was not the case when MHA1L was co-expressed with ACD6 L300F , which has a substitution in the ankyrin domain ( Figures 5C and 5D). To verify the observation in the oocyte system, we transiently expressed standard ACD6 from Col-0, the gain-of-function ACD6-1 variant or the ACD6 L557F variant in human embryonic kidney 293 (HEK293) cells. Codon optimization was required for efficient expression of ACD6 ("opti2-ACD6") in HEK293 cells ( Figures S5E and S5F). opti2-ACD6-mGFP5-6xHis and MHA1-L-mGFP5-6xHis were both observed at the plasma membrane ( Figure S5F), with stronger GFP signal enriched at the plasma membrane of dying cells ( Figure S5G). Calcium influx induced by increasing extracellular calcium, from 0.1 mM to 2.5 mM, was monitored using the ratiometric fluorescent dye Fura-2, which binds to intracellular calcium 23,40-42 . Compared to cells transformed with an empty vector, calcium influx was increased in cells expressing opti2-ACD6-myc-6xHis from Col-0 or the gain-of-function variant opti2-ACD6-1-myc-6xHis, but not in cells expressing the opti2-ACD6 L557F -myc-6xHis loss-of-function variant (Figures 5E and 5F). Co-expression of MHA1L-myc-6xHis enhanced the calcium influx in response to elevation of extracellular calcium, recapitulating the observations from oocytes ( Figure 5G). Together, these results show that ACD6 can stimulate ion channel activity in two different heterologous systems, Xenopus oocytes and human HEK293 cells, and that this activity can be further enhanced by MHA1L.

MHA1L and ACD6 are required for calcium signaling in PTI responses
To obtain in planta evidence for ACD6 being involved in calcium influx, we introduced the luminescent aequorin calcium biosensor, which has been used to assay responses to pathogen-and damage-associated molecular patterns (PAMPs/DAMPs) 43-45 , into acd6-ko, acd6-1 and MHA1L overexpressor backgrounds, using the calcium channel mutant death, no defense1 (dnd1-1) as a positive control 20 . Compared to wild-type plants, flg22-induced calcium influx was greatly reduced in acd6-ko loss-of-function mutants ( Figure 6A). We confirmed this finding with the R-GECO1 calcium reporter, which has also been used to monitor flg22-induced calcium fluxes 46 ( Figure S6A). Similarly, loss-of-function of MHA1L impaired calcium influx triggered by flg22 treatment (Figure 6A), which is consistent with impaired flg22-triggered ROS burst and MAPK activation in mha1l mutants (Figures 2A and 2B). These were specific defects, since the PAMP chitin and the stressor NaCl did not affect calcium influxes in either acd6-ko or mha1l mutants (Figures 6B and 6C). Similar specificity was observed when ACD6-Est-1 was knocked out in the Est-1 background (Figures S6C-S6F).
Together with the observation that MHA1L and ACD6 can stimulate calcium channel activity in Xenopus oocytes and human HEK293 cells, we propose that ACD6 contributes to the regulation of calcium influx, which can be directly activated by MHA1L during plant immunity.
Since ACD6 could stimulate calcium influx in both animal and plant cells, we further tested whether calcium is required for ACD6-triggered immunity. To this end, we cultured the gain-of-function mutant acd6-1 in a hydroponic system that contained either 1.5 or 0.1 mM external calcium. Similar to the calcium channel mutant dnd1-1 20 , we found that depletion of calcium attenuated signs of autoimmunity and expression of PR1 in acd6-1 ( Figure 6D). The ankyrin repeats of ACD6 are similar in sequence to those of mammalian TRP proteins, and similar to certain TRP proteins, ACD6 functions can stimulate ion flux across membranes ( Figure S5). There is extensive knowledge on the structure and function of TRP proteins, providing potential models for understanding the activity of ACD6 [31][32][33] . However, the transmembrane domains of ACD6 and TRP channels differ, and it is therefore unlikely that ACD6 and homologs function in exactly the same manner as TRP proteins do. Similar to a subset of TRP proteins, ACD6 can apparently be regulated through its ankyrin repeats by small ligands, the MHA1 and MHA1L proteins, with MHA1L enhancing the accumulation and activity of both the standard ACD6 protein and its experimentally induced derivative ACD6-1, but not of the natural ACD6-Est-1 variant. In contrast, MHA1-Ty-0, but not the standard MHA1 protein, can suppress activity of the ACD6-Est-1 variant (Figures 6E and S3G).
TRP channels typically function as tetramers, with two membrane-spanning helices from each subunit contributing to the ion channel pore (see Ref. 33 for a recent compilation of TRP channel structures). ACD6 is also found in a large complex 11,16 (Figures 4H and 4I), but the size of this complex is substantially greater than that of a simple tetramer, suggesting either a higher oligomeric state or association with additional proteins. Several TRP channels can form heteromultimers, with heteromultimers often having different activities than the homomultimers 31 . In A. thaliana, combinations of natural ACD6 alleles, including those from the Mir-0 and Se-0 accessions, can lead to increased ACD6 activity in F1 hybrids relative to either parent 2,3 , compatible with a scenario in which the variants assemble into heteromultimers in the F1 hybrids that have different properties than the respective homomultimers.
Also similar to ACD6, gain-of-function mutations in TRP channel genes are common, with many of these having been identified as variants underlying human genetic diseases 47 . ACD6 itself was originally discovered based on the experimentally induced gain-of-function, hyperactive acd6-1 allele 4 . A highly active natural allele, ACD6-Est-1, was subsequently found to be common in natural populations of A. thaliana 1 . The Mir-0/Se-0 combination described above can also be considered as a gain-of-function situation. In all four alleles, the sequences responsible for increased activity are found in the predicted transmembrane portion of ACD6 ( Figure S6G). While some gain-of-function mutations in TRP channels affect either the N-terminal portions including the ankyrin repeats or the C-terminal portions, many alter amino acids in the transmembrane domain (Table S1). This is significant, because, as in ACD6, the overall length of the transmembrane domains is typically much shorter than the length of the other domains, making for a smaller mutational target.
TRP channels can integrate inputs consisting of different stimuli or ligands, and the opposite effects of MHA1 and MHA1L on ACD6 activity are reminiscent of the effects of different ligands on TRPV1 activity 48 .
The exact role of MHA1 in regulating ACD6 remains somewhat enigmatic, because the clearest effects are exerted by a natural dominant-negative allele that attenuates the activity of the highly active ACD6-Est-1 allele.
It is conceivable that MHA1 is primarily a regulator of an ACD6 homolog; A. thaliana genomes encode about half a dozen proteins with close similarity to ACD6 in both the ankyrin repeats and transmembrane domains.
Redundancy among these proteins may also explain why inactivation of ACD6 has relatively minor consequences 10 compared to the pronounced phenotypic effects resulting from ACD6 hyperactivity or MHA1L overexpression (refs. [1][2][3][4] and this work). On the other hand, the effects of MHA1L overexpression seem to depend entirely on ACD6, pointing to an exclusive relationship between MHA1L and ACD6. It may appear surprising that MHA1L enhances accumulation of an ACD6 complex, rather than merely regulating its activity, but increased accumulation of ACD6 has also been observed in acd6-1 mutants 11 , consistent with positive feedback regulation of ACD6. One possibility is that MHA1L affects ACD6 activity at least in part by stabilizing ACD6, which has been shown to be regulated by protein degradation in the cytoplasm 11 , via other membraneresident proteins with which ACD6 associates 11,12,16 .
Although the ankyrin repeats of TRP proteins are similar to those of ACD6 and related plant proteins, it is unlikely that the ACD6 transmembrane domains adopt a topology similar to those of TRP channels and it is therefore not obvious from the secondary structure whether ACD6 acts as an ion channel ( Figure   S5C). Nevertheless, ACD6 can stimulate calcium influx both in two heterologous systems, Xenopus oocytes and human cell culture, and in plants, and several ACD6 effects in planta appear to require calcium ( Figures   5 and 6). Calcium influx from extracellular stores is an early event in the host response to pathogens, with relatively well characterized downstream steps. One of the earliest cellular responses to PAMPs is the rapid increase of cytoplasmic calcium levels, indicating that calcium channels are intimately associated with PTI. In agreement, genetics has implicated cyclic nucleotide-gated channel (CNGCs) in this process, and at least one of these CNGCs is a functional calcium channel whose activity is regulated by PAMPs/DAMPs, such as bacterial lipopolysaccharide (LPS) and flg22 or plant elicitor peptide 3 (Pep3) 20,36-38,49,50 . PAMPs also regulate activity of calcium channels from the OSCA family, which are important in stomata-dependent immunity 21 .
Similar to the experimentally induced acd6-1 and the natural ACD6-Est-1 alleles, mutations in genes for several plant calcium channels and transporters, including the ACA1 and ACA4 calcium-ATPases, the CAX1 and CAX3 vacuolar H + /calcium transporters, and the CNGC2/DND1 and CNGC4/DND2 calcium channels, can cause autoimmunity 20,51-53 . We propose that ACD6 modifies PAMP-triggered calcium signals, supported by the finding that the calcium environment affects both autoimmunity in acd6-1 gain-of-function mutants and the proliferation of PTI-inducing Pst DC3000 hrcCbacteria in acd6-2 loss-of-function mutants ( Figure 2C). Calcium also plays a role in ETI, as cngc2/dnd1 mutants are impaired in the response to both avirulent and virulent bacterial pathogens 51 . Moreover, calcium channel blockers suppress cell death activated by the NLRs RPM1 and ZAR1 54,55 , and ETI has recently been linked directly to calcium influx through the discovery that diverse NLRs and hNLRs are calcium permeable channels themselves 22,23 . ACD6 in turn has been genetically linked to the NLR SNC1 17 and to helper NLRs of the ADR1 and NRG1 families ( Figure   S3C). Our discovery of the MHA1/MHA1L family of peptides strengthens the case for ACD6 being a dynamic immune regulator associated with calcium influx (Figure S6G), since MHA1L, which can activate strong, ACD6-dependent immune responses, appears to be transcriptionally induced upon pathogen infection ( Figure S2A). In addition, it has been reported that ACD6 and FLS2 form complexes in planta, with plasma membrane association of both FLS2 and BAK1 in response to SA signaling being enhanced by ACD6 11 . These observations are in agreement with our finding that ACD6 has an important role in PTI, regulating calcium influx in response to the PAMP flg22.
Several other ACD6-related transmembrane domain proteins with ankyrin repeats are involved in plant immunity. Cereals encode a series of ACD6-related proteins 56 . One of these has been implicated in resistance to the smut fungus Ustilago maydis in maize 14 , while another one, Lr14a, has been shown directly to confer rust resistance in wheat 15 . Many of the residues that are required for ACD6 function are conserved in Lr14a ( Figure 3B). In A. thaliana, BDA1, predicted to have four instead of five transmembrane segments as in ACD6, is required for activity of the receptor-like protein SUPPRESSOR OF NPR1-1, CONSTITUTIVE 2 (SNC2) in plant immunity 13 . As with the induced acd6-1 gain-of-function allele and the naturally highly active ACD6 alleles, a mutation causal for BDA1 gain-of-function activity maps to the transmembrane domain.
Whether any of these ACD6 homologs depend at least in part on small peptide ligands remains to be determined.
In summary, our results demonstrate a complex set of relationships between different alleles and paralogs in the ACD6/MHA1/MHA1L system of small peptide-modulated calcium influx during pathogen responses, as well as positive and negative gain-of-function activities. Further complexity is added by the aggregate activity in this system being either temperature-sensitive or -insensitive, depending on the MHA1Lrequirement of the ACD6 allele ( Figure S3G). Our findings thus illustrate once more how naturally evolved special alleles, which are unlikely to be recoverable from conventional mutant screens, can provide new insights into fundamental aspects of biology.

Plant material and growth conditions
Arabidopsis thaliana accessions and Nicotiana benthamiana were derived from stocks maintained in the lab. Seeds were germinated and cultivated in growth rooms at a constant temperature of 23°C or 16ºC, air humidity at 65%, 16 h (long days) or 8 h (short days) 110 to 140 µmol m -2 s -1 light provided by Philips GreenPower TLED modules (Philips Lighting GmbH, Hamburg, Germany) with a mixture of 2:1 DR/W LB (deep red/white mixture with ca. 15% blue) and W HB (white with ca. 25% blue), respectively.
The hydroponic system used for assessing the effects of calcium has been described 67

Genome-wide association study (GWAS)
Severity of cell death (leaf necrosis) was scored on an arbitrary scale from 1 to 5 using six biological replicates as described 17 , and GWAS with efficient mixed-model (EMMAX) methods was performed with the easyGWAS web interface 61 . The Bonferroni correction with a threshold of 0.05 for multiple testing corresponded to an uncorrected p-value of 7.6x10 -8 . The variance explained (adjusted R 2 ) by the two loci was estimated using a Generalized Linear Model (GLM) using R 68 , with cell death as response variable and the two SNPs targeting the MHA loci as fixed effects with an interaction term.

Linkage disequilibrium calculation
Genomic regions surrounding MHA1 and MHA2 were subset from a short read VCF of 1001 Genomes data 69 using vcftools version 15.1 (ref. 70 ). Linkage disequilibrium R 2 values were calculated with PLINK version 1.9, with a window of 15 kb and an R 2 threshold of 0. for selection of seeds with or without the transgene. Target regions were PCR amplified using oligonucleotide primers in Table S2.

qRT-PCR
RNA was extracted from at least three biological replicates of pooled seedlings using the RNeasy kit (Thermo  Table S2.

Transgenic lines
An ACD6 fragment corresponding to Chr4: 8292084..8298360 in the TAIR10 reference genome was amplified

Trypan Blue staining
Freshly harvested leaf tissue was stained by completely immersing it in lacto-phenol/Trypan Blue staining solution (10 ml lactic acid, 10 ml glycerol, 10 ml phenol, 10 mg Trypan Blue and 10 ml water) and heating in a heat block at 80°C for 1 hour. Staining solution was aspirated and replaced by chloral hydrate solution (2.5 g/ml) to destain and clear the tissue. This was repeated once overnight to improve clearing. Samples were kept in 60% (v/v) aqueous glycerol for storage and further imaging.

Measurement of ROS production
Leaf discs (

Pathogen infection
Flood-inoculation and syringe-inoculation were used in this study. For flood-inoculation, three-week-old seedlings grown on ½ MS plate in short days were inoculated by Pst DC3000 or Pst hrcCas described 80 .
Bacteria grown on King's B (KB) media plate were harvested and resuspended in sterile distilled water to a final concentration of OD600=0.02. The bacterial suspension was dispensed into plates containing A. thaliana seedlings, and plates were incubated for 3 min before bacteria were removed by decantation.The entire rosette of each inoculated seedling was collected 3 days post-inoculation (dpi), and their fresh weight was measured before surface-sterilization (5% H2O2 for 3 min, followed by 3 washes with sterile distilled water) for colony assays. For syringe-inoculation, Pst DC3000 avrRpt2 or Pst DC3000 avrRps4 were harvested from KB plates and resuspended in 10 mM MgCl2 to a final concentration of OD600=0.0005. The suspension was infiltrated into leaves of 4-week-old seedlings with a needleless syringe. Bacterial growth was determined at 3dpi by colony counting.

Confocal microscopy
Five-day-old A. thaliana seedlings were imaged on a TCS SP8 confocal microscope (Leica, Wetzlar, Germany) with a 40x water corrected objective (1.10 NA) and a 488 nm excitation laser at 5% intensity. GFP emission was captured from 499 to 559 nm with a photomultiplier tube, at a gain of 450.3. Propidium Iodide emission was captured from 576 to 691 nm with a Hybrid detector, at a gain of 55.3. For the BFA assay, the laser intensity was reduced to 2%. GFP emission was captured from 507 to 539 nm with a Hybrid detector, at a gain of 10. FM4-64 emission was captured from 705 to 755 nm with a Hybrid detector, at a gain of 50. Images were combined to a frame average of 4. HEK293 cells transfected with plasmid DNA (pcDNA3.2, CMV::GFP, CMV:: opti2-ACD6 Col-0 -mGFP5-6xHis, or CMV::MHAL-mGFP5-6xHis) and incubated for 24 h were visualized through Carl Zeiss 880 Confocal microscopy system (Carl-zeiss, Oberkochen, Germany) at 20X magnification and a 488 nm excitation laser at 5% intensity. GFP emission was captured from 490 to 543 nm with a photomultiplier tube at a gain of 500 (for GFP) or 1000 (for ACD6 Col-0 -mGFP5-6xHis and MHAL-mGFP5-6xHis).

Subcellular fractionation
We followed a published protocol 81

Injection of oocytes and electrophysiological measurements
Dissected and preselected Xenopus laevis oocytes were obtained from Ecocyte Bioscience (Dortmund, Germany  was increased to 2.5 mM by adding a standard buffer supplemented with Ca 2+ into the chamber. Additional images were collected for the next 305 s.
The genotyping primers for the aequorin transgene insertion, acd6-ko, and acd6-1 are listed in Table S2. The