lines ofN. attenuataplants individually RNA
interference (RNAi)–silenced [inverted repeat
(ir)] or overexpressed (ov) in different JAZ
genes andNaMYC2to evaluate JA signaling–
deficiency; inNaMYB8, the phenolamide
master TF regulator; and inDH29andCV86,
which catalyze spermidine conjugation steps
in phenolamide biosynthesis, were screened
in a glasshouse open-choice screening exper-
iment using laboratory colonies ofEmpoasca
decipiens(Fig.2Aandfigs.S4andS5).Nymphs
and adultE. decipienspreferentially selected
irMYC2, irMYB8, and ovJAZiplants for feed-
ing and reproduction in contrast to the other
transgenic lines, which were only slightly dam-
aged by a few probing events (Fig. 2A). Diverse
JAZ proteins allow the JA signaling cascade to
regulate an array of metabolic and develop-
mental traits in different tissues at different
times ( 18 , 19 ), thereby contextualizing responses
and optimizing fitness. The modularity of the
JA-JAZi sector may provide specific responses
relevant toEmpoascaleafhoppers. A tissue-
wide transcriptomics analysis of all JAZ genes
in theN. attenuatagenome revealed that
NaJAZiis highly expressed in flower tissues
and is not responsive toM. sextaattack in
leaves ( 20 ), whereasNaJAZhshowed the op-
posite pattern (Fig. 2B and fig. S6). These pat-
terns were confirmed by exposing leaves to
E. decipiensandM. sextaattack and moni-
toring the kinetics of the JAZ transcript accu-
mulations:M. sextafeeding elicitedNaJAZh
transcripts in leaves, whereasEmpoascafeed-
ing elicitedNaJAZitranscript in leaves (Fig.
2B). Yeast two-hybrid (Y2H) assays revealed
that NaJAZi interacts with NaMYC2a, whereas
NaMYB8 interacts with NaMYC2b (fig. S7).
These data reveal that a sector of JA signaling
involving MYC2, MYB8, and JAZi is engaged
inEmpoascaresistance in leaves.
To identify the metabolites elicited by this
JA sector, we reared eitherE. decipiensadults
and nymphs orM. sextalarvae on leaves of
rosette-stage plants of JA signaling–deficient
transgenic lines (Fig. 2A) as well as on irAOC
and irCOI1plants; on irGGPPS, which are de-
ficient in 17-HGL-DTGs accumulations; and on
irPMTplants, which are impaired in nicotine
accumulations (fig. S8). We used an analyti-
cal and computational workflow ( 14 , 15 , 21 )
to collect high-resolution indiscriminant (data-
independent) tandem mass spectrometry (MS/
MS) spectra (termed idMS/MS) from extracts
ofEmpoasca- andManduca-damaged leaves.
We quantified metabolome specialization (dj
index), metabolome diversity (Hj index), and
metabolic specificity of individual metabolites
(Si index) using an information theory frame-
work ( 21 , 22 ). In the dimensions of information
theory–processed metabolome specialization
and diversity,M. sextaattack elicited overall
higher metabolome plasticity, resulting in
higherdj scores than those elicited by attack
byE. decipiens. The different transgenic lines
showed distinct trajectories of metabolome
plasticity reprogrammed by the attack of the
two insect species (Fig. 2C and fig. S9).
Focusing on transgenic lines preferred by
Empoasca, we noticed that the distinct signa-
tures of metabolome specialization elicited
byherbivoreattackwereweakerinirMYC2
and irMYB8plants (Fig. 2C and fig. S9). This
suggested that MYC2 is a master regulator of
metabolome plasticity in response to insect
attack and that MYB8-dependent herbivory-
induced phenolamides make up the metabolic
sector responsible for the increases in metab-
olome specialization. The separation of the
trajectories of metabolome changes elicited by
EmpoascaandManducaattack in the ovJAZi
lines, rather than their abolishment, further
pointed to a small set of metabolites elicited
byEmpoascaattack, regulated byNaJAZi, and
potentially involved inEmpoascaresistance.
To identify these metabolites, we ranked Si
scores for metabolite specificity calculated for
each MS/MS spectrum from theE. decipiens–
elicited metabolomes from the four transgenic
lines and linked the Si scores with coexpres-
sion heatmaps derived from correlations cal-
culated among individual metabolites and
Empoascanumbers and damage using the
global variance generated from all reverse-
genetics lines used in the feeding experiment
(Fig. 2D and fig. S9). Phenolamides ranked at
the top of the metabolic specificity Si scores,
with the putrescine-derived metabolites among
the highest, and these were negatively corre-
lated withEmpoascanumbers and damage, in
contrast to particular 17-HGL-DTGs, quinate
conjugates, and nicotinic acid, which showed
positive correlations (Fig. 2D).
The putrescine-derived phenolamides, CoP,
CP, and FP, were reduced in irAOC, irCOI1,
irMYC2, and irMYB8lines and selectively de-
creased in ovJAZiplants damaged byEmpoasca
feeding but not in those damaged byManduca
feeding, whereas other spermidine-derived
phenolamides showed similar responses to
the attacks of the two herbivore species in
ovJAZiplants (fig. S10). The spermidine-derived
metabolites could be excluded as mediators
ofEmpoascanonhost resistance on the basis
of the lack of responses of leafhoppers to the
irDH29and irCV86lines (Fig. 2A). To further
explore the involvement of CoP, CP, and FP, we
conducted in vivoEmpoascachoice assays by
individually infiltrating physiologically rele-
vant concentrations of synthetic CoP (7mM), CP
(100mM), and FP (10mM) into leaves of irMYC2
plants that are devoid of elicited phenolamides
(Fig. 2E and fig. S11). However, these infiltra-
tions did not alter the preference ofEmpoasca
for irMYC2plants. In vitroEmpoascadirect
feeding assays conducted with individual
compounds at physiologically relevant con-
centrations in glucose solutions revealed nosignificant changes in mortality rates of
E. decipienscompared with those fed on
glucose controls (Fig. 2E). These data suggest
that CoP, CP, and FP were not directly respon-
sible forEmpoascaresistance and that other
yet-unknown putrescine-derived phenolamide
metabolites were responsible.Multi-omics reveals the defense and its
three-pronged pathway
Leaves of herbivore-attackedN. attenuata
plants grown in the glasshouse accumulate a
variety of putrescine- and spermidine-derived
phenolamides ( 14 , 16 ). We selected 15 RILs
from the field-based multi-omics dataset of
the MAGIC population (Fig. 1) that accu-
mulated high levels of structurally diverse
OS-induced phenolamides to construct idMS/
MS and identify the structures of putrescine-
derived phenolamides. This effort resulted in
518 nonredundant idMS/MS spectra (Fig. 3A).
We performed a biclustering analysis to cluster
spectra according to fragment [normalized dot
product (NDP)] and neutral loss (NL)–based
similarities, which resulted in seven modules
(Fig. 3A). Module 5, particularly enriched in
phenolamide-related compounds containing
caffeoyl or putrescine moieties, was further
mapped onto a molecular network (Fig. 3A).
An unknown compound at mass/charge ratio
(m/z) 347.196 ([M+H]+,C 19 H 27 N 2 O 4 +) occupied
the first layer of directly linked network neigh-
bors for the two isomers of CP (m/z251.14)
because of their shared neutral losses of pu-
trescine ofD88.10 Da and fragment peak at
m/z163.04 (C 9 H 7 O 3 +) corresponding to the
caffeoyl moiety (Fig. 3A and fig. S12). The
idMS/MS for a fragment peak atm/z259.09
(C 15 H 15 O 4 +), which resulted from the loss of
putrescine of the molecular ion, further frag-
mented tom/z163.04 with a neutral loss of
96.055Da(C 6 H 8 O). This implied that the un-
knownm/z347.19 is a CP derivative decorated
with a C 6 H 8 O residue on the aromatic ring
of the caffeoyl moiety (fig. S12), which had
previously been associated with JA signaling
in natural accessions ofN. attenuata( 14 ).
To test whether the unknownm/z347.19
metabolite is regulated by the specific JA-JAZi
module, we explored the coassociation network
and conducted coexpression analyses for in-
ducedm/z347.19 againstEmpoascanum-
bers and damage and JAs in the field-planted
MAGIC population (Fig. 1). Them/z347.19 was
negatively correlated withEmpoascadamage
but positively correlated with JA, JA-Ile, and
JA-Val (fig. S13). We then mined theEmpoasca-
induced metabolomes of JA-deficient transgenic
lines (Fig. 2). However, with similar computa-
tional workflows, we were unable to identify
this compound in the dataset described earlier
(Fig. 2). Extensive experimentation revealed
that field-planted RILs elicited more of this
unknown compound than glasshouse-grownBaiet al.,Science 375 , eabm2948 (2022) 4 February 2022 3of9
RESEARCH | RESEARCH ARTICLE