H
OH metabolites
OH
Article
Metabolomic Abnormalities in Serum from Untreated and
Treated Dogs with Hyper- and Hypoadrenocorticism
Carolin Anna Imbery 1,2 , Frank Dieterle 3,†, Claudia Ottka 4,5,6,† , Corinna Weber 2 , Götz Schlotterbeck 3,
Elisabeth Müller 2, Hannes Lohi 4,5,6 and Urs Giger 1,7,*
1 Vetsuisse Faculty, University of Zürich, 8057 Zürich, Switzerland; carolinanna.imbery@uzh.ch
2 Laboklin GmbH & Co. KG, 97688 Bayern, Germany; weber@laboklin.com (C.W.);
mueller@laboklin.com (E.M.)
3 Institute for Chemistry and Bioanalytics, School of Life Sciences, University of Applied Sciences
Northwestern Switzerland, 4132 Muttenz, Switzerland; fd@frank-dieterle.de (F.D.);
goetz.schlotterbeck@fhnw.ch (G.S.)
4 PetMeta Labs Oy, 00300 Helsinki, Finland; claudia.ottka@petmetalabs.com (C.O.);
hannes.lohi@petmetalabs.com (H.L.)
5 Department of Veterinary Biosciences and Department of Medical and Clinical Genetics,
University of Helsinki, 00100 Helsinki, Finland
6 Folkhälsan Research Center, 00250 Helsinki, Finland
7 Section of Medical Genetics, University of Pennsylvania, Philadelphia, PA 19104, USA
* Correspondence: giger@upenn.edu
† These authors contributed equally to this work.
Abstract: The adrenal glands play a major role in metabolic processes, and both excess and insufficient
serum cortisol concentrations can lead to serious metabolic consequences. Hyper- and hypoadreno-



 corticism represent a diagnostic and therapeutic challenge. Serum samples from dogs with untreated


hyperadrenocorticism (n = 27), hyperadrenocorticism undergoing treatment (n = 28), as well as
Citation: Imbery, C.A.; Dieterle, F.;
Ottka, C.; Weber, C.; Schlotterbeck, G.; with untreated (n = 35) and treated hypoadrenocorticism (n = 23) were analyzed and compared to
Müller, E.; Lohi, H.; Giger, U. apparently healthy dogs (n = 40). A validated targeted proton nuclear magnetic resonance (
1H NMR)
Metabolomic Abnormalities in Serum platform was used to quantify 123 parameters. Principal component analysis separated the untreated
from Untreated and Treated Dogs with endocrinopathies. The serum samples of dogs with untreated endocrinopathies showed various
Hyper- and Hypoadrenocorticism. metabolic abnormalities with often contrasting results particularly in serum concentrations of fatty
Metabolites 2022, 12, 339. https:// acids, and high- and low-density lipoproteins and their constituents, which were predominantly
doi.org/10.3390/metabo12040339 increased in hyperadrenocorticism and decreased in hypoadrenocorticism, while amino acid con-
Academic Editors: Silvia Ravera and centrations changed in various directions. Many observed serum metabolic abnormalities tended
Markus R. Meyer to normalize with medical treatment, but normalization was incomplete when compared to levels
in apparently healthy dogs. Application of machine learning models based on the metabolomics
Received: 18 February 2022
data showed good classification, with misclassifications primarily observed in treated groups. Char-
Accepted: 6 April 2022
acterization of metabolic changes enhances our understanding of these endocrinopathies. Further
Published: 9 April 2022
assessment of the recognized incomplete reversal of metabolic alterations during medical treatment
Publisher’s Note: MDPI stays neutral may improve disease management.
with regard to jurisdictional claims in
published maps and institutional affil- Keywords: Cushing’s syndrome; Morbus Addison; canine; nuclear magnetic resonance; laboratory
iations. diagnostics; endocrinopathy
Copyright: © 2022 by the authors.
Licensee MDPI, Basel, Switzerland. 1. Introduction
This article is an open access article Gluco- and mineralocorticoids, synthesized by the adrenal cortex, play an important
distributed under the terms and role in homeostasis of glucose, protein, and fat metabolism, enabling an appropriate stress
conditions of the Creative Commons response, and maintaining blood pressure and electrolyte balance [1,2]. Corticosteroid im-
Attribution (CC BY) license (https:// balances can lead to serious health problems in humans and animals, including dogs [3–6].
creativecommons.org/licenses/by/ Hyperadrenocorticism or Cushing’s syndrome reflects a chronic excess of glucocorticoids
4.0/).
Metabolites 2022, 12, 339. https://doi.org/10.3390/metabo12040339 https://www.mdpi.com/journal/metabolites
Metabolites 2022, 12, 339 2 of 21
caused by adrenal cortex or pituitary neoplasia or can develop iatrogenically by administra-
tion of glucocorticoids [6,7]. Varied clinical signs are associated with hyperadrenocorticism
and hormonal and imaging tests are applied diagnostically [6,7]. Depending on the cause,
the treatment may involve surgical intervention or medical treatment [6,7] with drugs such
as the synthetic steroidogenesis inhibitor trilostane, which is frequently used in dogs [7].
In contrast, hypoadrenocorticism or adrenal insufficiency or Morbus Addison refers
to a deficiency of glucocorticoids with or without lack of mineralocorticoids that results
from various defects in the adrenal axis [5,8]. The clinical signs of hypoadrenocorticism
vary greatly from mild unspecific signs to life-threatening adrenal crisis, and the diagnosis
is based on hormonal testing [5,8]. Hormonal replacement is the mainstay of long-term
treatment, while supportive therapy is necessary for cases of emergency [5,8,9].
Specific hormonal and metabolic changes have been investigated in both
endocrinopathies [5–8,10]. However, comprehensive assessments of the global serum
metabolomes of patients suffering from hyper- or hypoadrenocorticism are rare in any
species [11–16].
Various technologies have been introduced to assess the metabolome in biological
samples, such as serum, including NMR spectroscopy and mass spectrometry (e.g., gas
chromatography–mass spectrometry (MS), liquid chromatography–MS) and enable the
identification and quantification of large numbers of metabolites [17,18].
Here, we applied a validated 1H NMR spectroscopy method optimized for dogs to
characterize the serum metabolomes of untreated and treated dogs with hyper- and hypoa-
drenocorticism. We hypothesize that (1) specific metabolic abnormalities will differentiate
between hyperadrenocorticism, hypoadrenocorticism, and control dogs, (2) machine learn-
ing models using solely the serum metabolomics data will correctly differentiate both
endocrinopathies and distinguish treated dogs from untreated and control dogs, and
(3) medical treatment of either endocrinopathy will lead to partial or complete reversal of
the metabolic abnormalities.
2. Results
2.1. Samples, Demographics, and Serum Cortisol Test Results
The serum samples selected fulfilled the entry criteria for the respective group of adult
dogs with either high cortisol concentrations by low-dose dexamethasone suppression tests
(LDDST) (hyperadrenocorticism untreated (HYPERU) group), or low cortisol concentra-
tions by adrenocorticotropic hormone stimulation tests (ACTH-ST) (hypoadrenocorticism
untreated (HYPOU) group), or were obtained from dogs with treated hyperadrenocorticism
with low cortisol concentrations by ACTH-ST (hyperadrenocorticism treated (HYPERT)
group; all unpaired samples with the HYPERU group) (Table 1), or with treated hypoad-
renocorticism (hypoadrenocorticism treated (HYPOT) group; 23 paired samples with the
HYPOU group), or from adult dogs with serum chemistry and complete blood count (CBC)
results within reference intervals and thus no laboratory evidence of diseases (control
(CONT) group).
A total of 153 left-over canine serum samples were included and analyzed by 1H NMR
spectroscopy, including 27 serum samples in the HYPERU, 28 in the HYPERT, 35 in the
HYPOU, 23 in the HYPOT, and 40 in the CONT group. Serum samples were mostly from
Germany (89%) and rarely from other European countries (Luxembourg, Czech Republic,
Romania, Finland, Norway, and Sweden). The dogs in both HYPER groups were signifi-
cantly older than the dogs in the HYPOU and CONT groups, albeit the age ranges were
large and overlapped (Table 1). The true effect of breed could not be statistically evalu-
ated, due to the low number of dogs per breed and group. Mixed breed dogs accounted
for the largest proportion enrolled in any group. The only breed overrepresented were
Dachshunds with seven and two dogs in the HYPERU and HYPERT groups, respectively.
No differences in sex or neutering status were observed between the groups (Table 1).
Metabolites 2022, 12, 339 3 of 21
Table 1. Demographic data of 153 serum samples from dogs in the groups of CONT (n = 40), HYPERU
(n = 27), HYPERT (n = 28), HYPOU (n = 35), and HYPOT (n = 23).
CONT HYPERU HYPERT HYPOU p Value HYPOU† HYPOT†
Number of dogs, n 40 27 28 35 23 23
Age, years, median (range) 5.3 (1.3–11.0) a 11.0 (8.0–14.0) b 11.0 (6.3–15.0) b 6.0 (0.8–12.0) a <0.001 6.0 (1.3–11.0) 6.0 (1.3–11.0)
Breeds, n
Mixed breed 11 10 11 18 11 11
Others ¥/Dachshund 29/0 10/7 15/2 17/0 12/0 12/0
Sex, n
Males, intact/castrated 12/9 8/7 9/4 9/12 4/10 4/10
Females, intact/spayed 10/8 ‡ 7/5 2/13 4/10 >0.05 1/8 1/8
Cortisol ACTH-ST, ng/mL, median (range)
Cortisol pre-ACTH ND ND 5.5 (1.4–13.8) 0.5 (0.5–4.3) 0.5 (0.5–4.3) ND
Cortisol post-ACTH ND ND 12.2 (3.0–19.9) 0.5 (0.5–6.1) 0.5 (0.5–3.9) ND
Cortisol LDDST, ng/mL, median (range)
Cortisol
pre-dexamethasone ND 55.9 (15.5–300.7) ND ND ND ND
Cortisol 8 hrs
post-dexamethasone ND 36.5 (11.2–86.4) ND ND ND ND
Note. ¥ Breed with ≤3 dogs/breed/group. ‡ Sex was not reported for one dog. † Comparison of the 23 paired
samples from dogs in HYPOU and HYPOT. Results with different letter superscripts (a, b) in the same line are
significantly different from each other. ACTH—adrenocorticotropic hormone; ACTH-ST—adrenocorticotropic
hormone stimulation test; CONT—control group; hrs—hours; HYPOU—hypoadrenocorticism untreated; HYPOT—
hypoadrenocorticism treated; HYPERU—hyperadrenocorticism untreated; HYPERT—hyperadrenocorticism
treated; LDDST—low-dose dexamethasone suppression test; ND—not determined.
2.2. Metabolomic Analyses
In the present study, all 105 serum metabolites assessed in the validation study for
canine samples [19] were identifiable and measurable by 1H NMR analysis. Furthermore,
11 relative concentrations of fatty acids and seven selected amino acid ratios were calculated.
The resulting 123 metabolic parameters were also documented in our recent study of canine
hepatopathies [20]. Medians for all parameters of the CONT group fell in the previously es-
tablished serum reference intervals for dogs of all ages [19]. However, for a few parameters,
25th or 75th percentiles of the CONT group fell slightly below (concentrations of citrate,
glutamine, glycoprotein acetyls (GlycA), and large very-low-density lipoprotein (L-VLDL)-
triglycerides) or slightly above (concentrations of glycine and high-density lipoprotein
(HDL) particle size) the published serum reference intervals (Supplementary Table S1) [19].
No metabolomic differences were observed between two age-dependent CONT sub-
groups (dogs < 6 years (yrs) old vs. dogs ≥ 6 yrs old) by univariate testing and by PCA of
their serum metabolomics data which demonstrated complete overlap between the two
clusters (Supplementary Figure S1). Thus, all control dogs were combined to one CONT
group in the subsequent bioinformatic analyses of the metabolomics data.
2.2.1. Metabolomic Comparison of HYPERU, HYPOU, and CONT Groups
Univariate testing of serum metabolomics data of the unpaired groups showed signifi-
cant differences in 108 of 123 parameters between dogs of the HYPERU, HYPOU, and/or
the CONT groups in post-hoc analyses (Supplementary Table S1). In the principal compo-
nent analysis (PCA) clustering was observed in canine serum samples from the HYPERU,
HYPOU, and CONT groups. While the tight cluster from the CONT group resided within
the other clusters, the broader clusters of the HYPERU and HYPOU groups were partially
distinct, and clusters extended in different directions (principal component (PC) 1 = 91.3%
and PC 2 = 4.9% of total variance; Figure 1a). Many variables were found to influence the
projection and separation of the groups, as reflected in marginalized parameters in the PCA
loadings plot, showing relative contributions and the relationships between the parameters
(Supplementary Figure S2a).
Metabolites 2022, 12, x FOR PEER REVIEW 4 of 22 
 
the CONT groups in post-hoc analyses (Supplementary Table S1). In the principal com-
ponent analysis (PCA) clustering was observed in canine serum samples from the 
HYPERU, HYPOU, and CONT groups. While the tight cluster from the CONT group re-
sided within the other clusters, the broader clusters of the HYPERU and HYPOU groups 
were partially distinct, and clusters extended in different directions (principal component 
(PC) 1 = 91.3% and PC 2 = 4.9% of total variance; Figure 1a). Many variables were found 
to influence the projection and separation of the groups, as reflected in marginalized pa-
Metabolites 2022, 12, 339 rameters in the PCA loadings plot, showing relative contributions and the relat4ioofn2s1hips 
between the parameters (Supplementary Figure S2a). 
 
FiguFreig 1u.r Me 1e.tMabeotalobmoloicm aicnaanlyaslyesse osfo sfesreurumm ssaammpplleess ffrroomm ddooggssi nint hthe eg rgoruopusposf ohfy hpeyrpaedrraednoreconroticcoisrmticism 
untreuantterdea (tHedY(PHEYRPUE, Rn =, 2n7=), h27y)p, ohaydproeandorecnoortciocritsimcis umnturnetarteeadte (dH(YHPYOPUO, n, =n 3=5)3, 5a)n, dan tdhet hceonctornotlr oglroup U U
(CONgrTo,u np =(C 4O0)N. (Ta,)n S=co4r0e)s.  (pal)oStcso orfe spprilnotcsipoaf lp croinmcippoanl ceonmt apnoanleynstias n(PalCysAis) (aPnCdA ()ba)n pda(rbti)apl alertaisatl sleqaustares–
discrsiqmuianraesn–td aisncarliymsiinsa (nPtLaSn-aDlyAsi)s b(PaLseSd-D oAn) mbaesteadboonlommeitcasb dolaotma iocfs sdeartuamof ssaermumplessa mofp dleosgosf dino gHsYinPERU 
(blueH),Y HPEYRPO(Ub l(uree)d,)H, aYnPdO CO(reNdT), (agnrdeeCnO)N gTro(ugrpese.n S)hgaroduepds .ciSrhcaledse drecpirrcelseesnret p9r5e%se ncto9n5f%idceonncfied inentecervals, U U
whilien tceorlvoarlse,dw dhoiltesc iollloursetdradtoet sinildluivstirdatueailn sdaivmidpulaelss. aTmhpel easx. eTsh earaex elsabareeleladb eblyed thbey tfhiresfit r(sptr(ipnrcinipciapla) lc) om-
ponecnotms pwonitehn ttshwe iptherthceenpteargceenst aogf evsaorfiavnarciea nocfe tohfet hdeadtaat aexepxplalainineedd bbyy tthhaattc ocommpopnoennetnint pina rpenatrheenstehse. ses. 
(c) T(hce)  T3hDe s3cDorsecso rpelsoptl ootf oPfLPSL-SD-DAA babsaesedd oonn mmeettaabboolloommicicssd dataatao fosfe sreurmumsa msapmleps loefsd oofg ds oingHs iYnP HERYPERU U
(blue(b),l uHe)Y, PHOYUP O(red(r)e,d a)n, adn Cd OCONNTT (g(grreeeenn)) ggrroouuppss. .( d(d) )V aVraiarbialebilme pimorptaonrcteanincep rionje pctrioojnec(VtiIoPn) s(cVoIrPes) osfcores U
of cocmompopnonenent t11 ooff tthhee PPLLSS--DDAAi iddeenntitfiyfiynigngth tehteo pto2p0 2d0is dcrisimcriinmatiinnagtpinagra pmaertaemrseintedres sicne nddeisncgenorddienrgo of rder 
of imimpoprotratannccee. .TThhee ccoolloorreeddl elgeegnedndon otnh ethrieg hrtiginhdt iicnatdeisctahteesr etlhaeti vreelaabtuivned aanbcuenodf athnecev aorfia tbhlees ,vwaritihables, 
withr eredda nadndb lubeluien dinicdatiicnagtihnigg hhaignhd alonwd vloalwue vs,arleusepse,c rtievseplye,cwtihvielelyb, ewigheiillelu bsetriagtee silnlueusttrraaltevsa lnueeus.tral val-
ues. 
To maximize the separation of the groups, partial least squares–discriminant analysis
(PTLoS m-DaAx)imwiazsea tphpel iseedp. aInratthieonP LoSf- tDhAe gmroduepl,st, hpeafirtrisatlt wleoascto smqpuoanrens–tsdcioscnrtrimibuinteadnt5 8a.n0%alysis 
(PLSo-fDthAe) twotasl  vaaprpialniecde .( cInom thpeo nPeLnSt-1D=A4 m7.1o%d,ecl,o tmhep ofinresnt ttw2 =o 1c0o.m9%p)o. nAesnftosr ctohnetPrCibAu,tethde r5e8.0% 
of thwea tsoctlauls vtearriniagnocfet h(ceothmrepeognreonutp 1s, =w 4it7h.1c%lus, tceorsmepxtoennednintg 2i n=t o10d.i9ff%er)e. nAt sd ifroerc ttiohnes P. ACgAa,i nth, ere 
the CONT group was most tightly clustered and overlapped with the broader clusters of
both diseased groups (Figure 1b). Adding the third component, which contributed 6.7%
of total variance, to create a 3D PLS-DA scores plot also showed tight clustering for the
 CONT group, while both clusters of the adrenal-diseased samples extended into different
dimensions (Figure 1c). The loadings plot for the PLS-DA model is shown in Supplementary
Figure S3. The PLS-DA model was validated by a 10-fold cross-validation with R2, Q2, and
Metabolites 2022, 12, 339 5 of 21
accuracy as displayed in Supplementary Figure S4. All figures show a robust model with
three components being selected as the optimal number of components based on the Q2
criterion. Furthermore, a permutation test with 2,000 permutations was performed, which
shows that the model is not overfitting the data (Supplementary Figure S5).
The top 20 metabolites that discriminated between the three groups were identified by
the variable importance in projection (VIP) scores of the first component of PLS-DA and
included many lipid-associated parameters, such as total, free, and esterified cholesterol,
various HDL-associated lipid fractions, and fatty acid concentrations (Figure 1d). The
first component of PLS-DA predominantly discriminated between HYPERU and HYPOU
groups, as those mainly varied on the x-axis.
Hierarchical cluster analysis of samples from dogs with untreated endocrinopathies
and the CONT group revealed three main clusters and excellent separation between the
groups with few exceptions. Some samples from the HYPOU and CONT groups overlapped
and were assigned to a cluster predominantly containing samples from the CONT group.
The samples in the HYPERU cluster were more distant, indicating a more different serum
metabolomic profile from the other two clusters (Figure 2).
 
Figure 2. Dendrogram of hierarchical cluster analysis of serum metabolomics data from canine
samples of either hyperadrenocorticism untreated (HYPERU, blue, n = 27), hypoadrenocorticism
untreated (HYPOU, red, n = 35), and control (CONT, green, n = 40) groups. Each number on the x-axis
reflects one serum sample. The y-axis shows the similarity levels expressed as Euclidean distances.
Horizontal and vertical lines depict clustering of samples and differences in the distances, respectively.
A hierarchical cluster heat map of the top 20 parameters from PLS-DA VIP scores
of the first component largely revealed higher serum metabolite concentrations in the
HYPERU group and lower metabolite concentrations in the HYPOU group. The hierarchical
cluster analysis of the heatmap assigned the samples from HYPERU and HYPOU groups
into two clusters with some exceptions. The samples from the CONT group did not form a
separate cluster but were rather distributed among those two clusters. Despite being split
by some CONT samples, two subclusters mainly consisting of HYPOU samples showed
similarity based on color intensity patterns (Figure 3).
1 
 
Metabolites 2022, 12, x FOR PEER REVIEW 6 of 22 
 
not form a separate cluster but were rather distributed among those two clusters. Despite 
Metabolites 2022, 12, 339 being split by some CONT samples, two subclusters mainly consisting of HYPOU sam6 opfl2e1s 
showed similarity based on color intensity patterns (Figure 3). 
 
FFiigguurree 33.. HHiieerraarrcchhiiccaall cclluusstteerr hheeaattmmaapp ((ffoorr ssaammpplleess aanndd vvaarriiaabblleess)) ooff sseerruumm mmeettaabboolloommiiccss ddaattaa ooff 
ccaanniinnee ssaammpplleess iinn hhyyppeerraaddrreennooccoorrttiicciissmm uunnttrreeaatteedd ((HHYYPPEERRU,, nn == 2277,, bblluuee)),, hhyyppooaaddrreennooccoorrttiicciissmm U
untreated (HYPOU, n = 35, red), and control (CONT, n = 40, green) groups. The top 20 parameters 
untreated (HYPO , n = 35, red), and control (CONT, n = 40, green) groups. The top 20 parameters
identified by partiUal least squares–discriminant analysis (PLS-DA) variable importance in projection 
i(dVeInPt)i fisecdorbeys poaf rctioaml lpeoasntesnqtu 1a rwese–rdei sucsreimd.i nEaanctha cnoalluymsisn( PreLpSr-eDsAen)tvs aorniaeb lseerimumpo sratamnpcelei nwpitrho jegcrtoiounp 
(mVIaPrk) isncgosr ecsoloofrecdo matp tohnee tnotp1. Twheer ecouloseredd.  lEeagcehndc oolnu mthne rreigphrte sinendtiscaotnese tsheer uremlatsiavme pmleetwabiothlitger coounp-
mceanrtkriantigosnsc owloitrhe ddiaffterthenet troepd. anTdh eblcuoel oinrteednslietgieesn idndoincatthinegr higighht ainnddi lcoawte svatlhueesr,e rleastipveectmiveetlayb. oHliotre-
ciozonncetanlt raantdio vnesrwticiathl bdliafcfekr leintesr eddepanicdt cblluseteirnitnegn soift iseasminpdleicsa atnindg phairgahmaentedrslo. w values, respectively.
Horizontal and vertical black lines depict clustering of samples and parameters.
Machine learning methods using solely the metabolomics data were capable of cor-
rectlyM calcahsisnifeylienagr nsianmgpmleest hinotdos euistihnegr soofl etlhyet huentmreeattaebdo leonmdioccsrdinaotpaawtheirees coarp atbhlee CofOcNorT- 
rgercotluypc lians smifyoisnt gcasasemsp (l7e8s–i8n8to%e, iSthueprpolfetmheenutnatrrye aTtaedbleen Sd2o)c. rTinhoupsa, twhiietsh otrhteh esiCmOpNleT loggroisutipc 
irnegmreossstiocnas meso(d7e8l–, 8888%%, oSfu pthpel esmamenptlaersy coTuabldle bSe2 a).ssTighnuesd,  wtoit thhteh ceosrirmecpt lgerloougpisst i(cTarebgleres s2- 
sainodn Sm3o, dSuelp, p8l8e%moenf ttahreys Eaqmupalteiosnc oSu1l)d. be assigned to the correct groups (Tables 2 and S3,
Supplementary Equation S1).
Table 2. Simple logistic regression model to classify dogs based on the metabolomics data into the 
Tgarboluep2s. oSfi m(ap) HleYloPgEiRstUic, HreYgPreOsUsi,o andm CodOeNl tTo, (cbla) sHsiYfyPdERogUs, HbaYsPeEdRoTn, athned mCOetNabTo, l(ocm) HicYs PdOatUa, iHntYoPtOheT, 
garnodu pCsOoNf (Ta )coHmYpPaErReUd ,tHo YthPeO cUli,naicnodpCatOhNolTo,g(ibc)alHlyY aPsEsRigUn,eHd YgProEuRpTs,.a nd CONT, (c) HYPOU, HYPOT,
and CONT compared to the clinicopathologically assigned groups.
 Groups Assigned by Simple Logistic Regression Clinicopathologically Dogs,  
Clinicopathologically Assigned Groups Dogs, n Groups Assigned by SimplMe Loodgeislt ic Regression Model
Assigned Groups n 
a a COCNOTNT HYHPYEPREUR HYPOU HYPOU U
CONT  CONT 40 40 38 38 1 1 1 1
HYPERU HYPERU 27 27 1 1 25 25 1 1
HYPOU  HYPO  35 35 6 6 2 2U 272 7
b b   COCNOTNT HYHPYEPREUR U HYHPYEPRETR T
CONT 40 40 0 0
 CONT 40 40 0 0 
HYPERU 27 1 24 2
HYPER  HYPERU 28 27 5 1 24 2 T 3 20
c  HYPERT 28 CONT5 H3Y PO H20Y U POT
CONT c  40  3C2ONT  HYPO3U HYPO5 T 
HYPOU  CONT 35 40 5 32 3 28 5 2
HYPOT  HYPOU 23 35 6 5 28 1 2 16
Note: CONT—control group; HYPOU—hypoadrenocorticism untreated; HYPOT—hypoadrenocorticism treated;
HYPERU—hyperadrenocorticism untreated; HYPERT—hyperadrenocorticism treated.
 
Metabolites 2022, 12, 339 7 of 21
Among the nine serum amino acids measured, phenylalanine concentrations were ele-
vated in both endocrinopathies (Figure 4a). The HYPERU group showed increased serum
concentrations of tyrosine, alanine, total branched-chain amino acids (BCAA), isoleucine,
and valine (Figure 4b, Supplementary Table S1), while histidine concentrations were only
elevated in the HYPOU group (Figure 4c). However, only slight changes in serum concen-
trations were observed for glycolytic metabolites, with lactate and pyruvate concentrations
slightly increased in the HYPERU group, and acetate and citrate concentrations slightly
Metabolites 2022, 12, x FOR PEER REVI
 i
EnWcr eased in both endocrinopathies. The concentrations of GlycA were markedly increase8 dof 22 
in the HYPERU group, but only slightly increased in the HYPOU group (Figure 4d).
 
FiFgiugurer e4.4 C. oCnocnecnetnrtartaiotinosn s(m(mmmolo/lL/)L o)fo pfhpehneynlayllaanlainnien e(a()a, )a,laanlainnien e(b()b, )h, ihstiisdtiidnien e(c()c, )g, lgylcyocporportoetieni nace-
tyalcse (tGyllsyc(GAl)y (cdA)), (hdig),hh-digehn-sdietyn sliitpyolpiprootperiontse i(nHsD(HLD) cLh)oclheostleersotelr (oel)(, eH),DHLD tLrigtrliygclyerciedriedse (sf)(,f )v,evreyr-ylo-w-
delonwsi-tdye lnipsiotyprloiptoeipnrso t(eVinLsD(LV)L cDhLo)lecshtoerleoslt (egro),l V(gL),DVLL tDriLgltyricgelryicdeersi d(hes),( ahn),da ntodtatol tfaalttfyat atyciadcsi d(is) (iin)  isnam-
plseasm fprolems  fdroomgs dino gtshein gtrhoeugprso uofp sCoOfNCTO, NHTY, PHEYRPUE, RHY, PHEYRPTE, RHY, PHOYUP,O an,da HndYHPOYTP. OTh.eT bhoexbeos xoefs the U T U T
HYPOU group are presented both from the unpaired group utilized in multivariate analyses (n = 35, 
of the HYPOU group are presented both from the unpaired group utilized in multivariate anal-left) and the paired group (n = 23, right) utilized in comparison of HYPOU and HYPOT groups. Boxes 
yses (n = 35, left) and the paired group (n = 23, right) utilized in comparison of HYPO and
indicate the lower to upper quartile (25th–75th percentile) and median value. Whiskers eUxtend to 
mHinYimPOuTmg arnodu pms.axBimoxuems  ivnadliuceaste. Othuetllioerws earreto shuopwpner aqs uinadrtiivleid(u2a5tl ho–p7e5nth ciprcelrecse onrti sleta) rasn. Dd amsheeddia lnines 
invdailcuaet.eW rehfiesrkeenrscee xitnetnedrvtoalms. inLiimneusm aabnodvem faixgiumruesm rveaflleucets .sOiguntilfiiecrasnatr edsihffoewrennacessi nbdeivtwideueanl osppeencific 
grcoirucpless (o* rp s<ta 0r.s0.5;D *a*s ph e<d 0.l0in1e; s**i*n pd i<c a0t.e00r1e)fe. rNenoctee: iCntOeNrvTal—s.coLninterosla gbroovuepfi; gHuYrePsOre—flehctypsiogandifirecU nanotcor-
ticdiisfmfer eunncetrsebaetetwd;e eHnYspPeOcTifi—chgyrpouopads r(e*npo<co0r.t0ic5i;s*m*  pt<re0a.t0e1d;; **H* Yp P<E0R.0U0—1)h. yNpeortea:drCeOnNocTo—rticcoinsmtro lun-
trgeraotuedp;; HYPPOEUR—T—hhypyopaedrarednroecnoorctiocristmiciusmnt rteraetaetde;dH. YPOT—hypoadrenocorticism treated; HYPERU—
hyperadrenocorticism untreated; HYPERT—hyperadrenocorticism treated.
2.2.2. Metabolomic Comparison of HYPERU, HYPERT, and CONT Groups 
As the metabolomic bioinformatic data analyses described above compared 
HYPERU, HYPOU, and CONT groups, we next compared the serum metabolomic patterns 
and abnormalities in untreated and treated dogs with hyperadrenocorticism. 
In both, PCA and PLS-DA scores plots, the broad clusters of the HYPERU group 
mostly entirely overlapped the clusters of the CONT group and the HYPERT group clus-
ters did so as well. However, even though clusters of the HYPERU and HYPERT groups 
were partly overlapping, they trended into different directions (Figure 5a,b). Total vari-
ance explained by PC 1 and 2 of the PCA model added up to 59.3%, very similar to the 
total variance contributed by the first two PLS-DA components, which added up to 58.2% 
(Figure 5a,b). According to the PCA loading plots, the separation of HYPERU and HYPERT 
groups is mostly due to elevated absolute fatty acid and lipid concentrations in the 
 
Metabolites 2022, 12, 339 8 of 21
Serum concentrations of total cholesterol (as well as concentrations of free and es-
terified cholesterol) were increased in the HYPERU group and decreased in the HYPOU
group, whereas total triglyceride concentrations were only increased in the HYPERU group
compared to the CONT group. Concentrations of HDL and small low-density lipoprotein
(LDL) particles and most of their associated lipids followed the pattern for total cholesterol
concentrations for both endocrinopathies (Figure 4e,f). Concentrations of large-LDL and
VLDL particles and most of their associated lipids were only increased in the HYPERU
group (Figure 4g,h).
Similarly, the absolute serum concentrations of total and specific fatty acids were
mostly increased in the HYPERU group and decreased in the HYPOU group (Figure 4i).
Among the relative fatty acid concentrations, palmitic acid was increased, while linoleic
acid was slightly decreased in the HYPOU group. Relative serum concentrations of
docosahexaenoic acid were decreased in HYPERU and increased in the HYPOU group
(Supplementary Table S1).
2.2.2. Metabolomic Comparison of HYPERU, HYPERT, and CONT Groups
As the metabolomic bioinformatic data analyses described above compared HYPERU,
HYPOU, and CONT groups, we next compared the serum metabolomic patterns and
abnormalities in untreated and treated dogs with hyperadrenocorticism.
In both, PCA and PLS-DA scores plots, the broad clusters of the HYPERU group mostly
entirely overlapped the clusters of the CONT group and the HYPERT group clusters did so
as well. However, even though clusters of the HYPERU and HYPERT groups were partly
overlapping, they trended into different directions (Figure 5a,b). Total variance explained
by PC 1 and 2 of the PCA model added up to 59.3%, very similar to the total variance
contributed by the first two PLS-DA components, which added up to 58.2% (Figure 5a,b).
According to the PCA loading plots, the separation of HYPERU and HYPERT groups is
mostly due to elevated absolute fatty acid and lipid concentrations in the HYPERU group,
while in the HYPERT groups relative concentrations of saturated fatty acids were increased,
in addition to other parameters (Supplementary Figure S2b).
The top 20 parameters identified by the first component of PLS-DA VIP scores were
mainly elevated in HYPERU group and low in the CONT group compared to the HYPERT
group. The most discriminating measurands included GlycA, phenylalanine, alanine, mul-
tiple fatty acids, several VLDL particles and associated lipids (Figure 5c). In the hierarchical
cluster analysis, samples from the HYPERU and CONT groups clearly separated into two
main clusters with only a few outliers, while the samples from the HYPERT group were
dispersed throughout the dendrogram (Figure 5d).
Various machine learning methods classified 70–88% of samples to the correct group
based on the serum metabolomics data alone (Supplementary Table S2). As such, the simple
logistic regression model was capable of assigning 88% of the samples correctly (Table 2b).
Most increased serum amino acid concentrations in the HYPERU group were normal-
ized in the HYPERT group (alanine, total BCAA, isoleucine, valine, and tyrosine), except
phenylalanine concentrations, which decreased but were still higher compared to the CONT
group (Figure 4a,b). Moreover, the serum histidine concentrations, which were unchanged
in HYPERU dogs, actually increased in the HYPERT group compared to the CONT group
(Figure 4c). Increased concentrations of lactate, pyruvate, and citrate in the HYPERU group
normalized in the HYPERT group, except for serum acetate concentrations which were still
elevated compared to the CONT group (Supplementary Table S1).
Metabolites 2022, 12, x FOR PEER REVIEW 9 of 22 
 
HYPERU group, while in the HYPERT groups relative concentrations of saturated fatty 
acids were increased, in addition to other parameters (Supplementary Figure S2b). 
The top 20 parameters identified by the first component of PLS-DA VIP scores were 
mainly elevated in HYPERU group and low in the CONT group compared to the HYPERT 
group. The most discriminating measurands included GlycA, phenylalanine, alanine, 
multiple fatty acids, several VLDL particles and associated lipids (Figure 5c). In the hier-
archical cluster analysis, samples from the HYPERU and CONT groups clearly separated 
Metabolites 2022, 12, 339 9 of 21
into two main clusters with only a few outliers, while the samples from the HYPERT group 
were dispersed throughout the dendrogram (Figure 5d). 
 
FFiigguurree 55.. MMeetatabboololommici canaanlaylsyesse sofo sfesreurmum samsapmlepsl efsrofmro mdogdso gins tihne tghreougprosu hpyspheryapderreandorceonroticcoisrtmi- 
cuinsmtreuatnetdre (aHteYdP(EHRYU,P nE =R 27, )n, h=yp2e7r),adhryepneorcaodrrteicnisomco trrteicaitsemd (tHreYatPeEdR(TH, nY =P E28R), ,annd= c2o8n)t,roaln (dCOcoNntTr,o nl  U T
(=C 4O0N). T(a, )n S=co4r0e)s.  (pal)oStsc oorfe psrpinloctispaolf cporminpciopnaelncto amnpaolynseins t(PanCaAly) sains d(P (CbA) p) aarntdia(l ble)apsat rstqiaulalreeass-tdsiqsucrairmesi--
nant analysis (PLS-DA) based on metabolomics data between serum samples of dogs in HYPERU 
discriminant analysis (PLS-DA) based on metabolomics data between serum samples of dogs in
(blue), HYPERT (green), and CONT groups (red). Shaded circles represent 95% confidence intervals, 
HwYhPilEe RcoUlo(brelude d),oHtsY iPllEuRstTra(tger einedn)iv, aidnudaCl sOaNmTplgerso. uTphse (arexdes). aSrhea ldabedelecidr cbleys trheep friersset natn9d5 %seccoonndfi d(pernicne-
icniptearlv) aclso,mwphoilneecnotlso rwedithd otthsei lpluesrtcreantetaingedsiv oidf uvaalrsiaanmcpel eosf. tThhee daaxteas aerxepllaabineleedd bbyy tthheatfi rcsotmanpdonseecnotn idn 
(pparrienncitphaesl)esc.o (mc)p Voanreinabtslew imithpothrteapnecerc ienn ptargoejescotifovna (rViaIPn)c escoofrtehs eodf caotamepxopnlaeinnte 1d obf ythteh aPtLcSo-mDApo indeennt-
itnifipeas rtehnet htoepse 2s0.  (dci)scVraimriainbaletinimg poartaamnceeteirns pinr odjescctieonndi(nVgIP o)rdsceorr oefs iomf pcormtapnocnee. nTth1e ocof ltohredP LleSg-eDnAd 
iodne nthtiefi reisghtht eintdoipca2te0s dthisec reimlaitnivaet ianbgupnadraanmce toefr vsairniadbelescs,e wnditihn greodr adnedr  obfluiem ipnodrictanticneg. hTighhe acnodlo lroewd  
lveagleuneds, ornestpheecrtiigvheltyi,n wdihcialtee sbethigeer eilllautsivtreataebsu nneduatnracle voaflvuaersi.a (bdle) sD, wenidthroregdraamn dofb lhuieerianrdchiciactailn cgluhsigtehr 
aanndallyoswis voafl useesru, rmes pmeecttaivbeollyo,mwihc ilreesbuelitgse firlloumst rcaatnesinnee usatrmalpvleasl uiens .e(idth)eDr eHndYrPoEgRraUm (bolfuhe)ie, rHarYcPhiEcRalT 
c(lgurseteenr)a, noarl ythseis CoOf sNerTu gmromuept a(breodlo).m Eiaccrhe snuultms fbreorm onca tnhien ex-saaxmisp rleefsleinctesi othneer sHerYuPmER sam(bplluee. )T, hHeY yP-EaRxis U T
shows the similarity levels expressed as Euclidean distances. Horizontal and vertical lines depict 
(green), or the CONT group (red). Each number on the x-axis reflects one serum sample. The y-axis
clustering of samples and differences in the distances, respectively. 
shows the similarity levels expressed as Euclidean distances. Horizontal and vertical lines depict
clustering of samples and differences in the distances, respectively.
Serum GlycA concentrations diminished in samples from the treated compared to the
HYPERU group but remained increased (Figure 4d).
While total cholesterol concentrations normalized in the HYPERT group, concentra-
tions of total triglycerides and most triglyceride subtypes remained elevated, except for
 large and small LDL triglyceride concentrations, which were not altered in either the
HYPERU or HYPERT groups. Similarly, concentrations of extra-large, large, and small
VLDL particles and associated lipids, with few exceptions, remained increased in the
HYPERT group compared to the CONT group. However, the concentrations of most
other lipoprotein particles and their associated lipid fractions (except triglyceride fractions)
decreased in the HYPERT group (Supplementary Table S1).
Metabolites 2022, 12, 339 10 of 21
Total and most individual fatty acid concentrations tended to decrease but some
remained elevated in the HYPERT group (Figure 4i). Relative concentrations of fatty acids
were mostly unchanged between HYPERU and HYPERT samples, except there was an
increased relative concentration of palmitic acid in the HYPERT compared to the HYPERU
group (Supplementary Table S1).
2.2.3. Metabolomic Comparison of HYPOU, HYPOT, and CONT Groups
The serum metabolomic patterns and abnormalities in untreated and treated dogs
with hypoadrenocorticism are compared below. However, while multivariate analyses
and the Kruskal–Wallis test were carried out using all collected HYPOU samples (n = 35),
for the Wilcoxon signed-rank test only the paired HYPOU and HYPOT samples (n = 23)
were included.
In the PCA, considerable overlap of HYPOU, HYPOT, and CONT group clusters was
observed (PC 1 = 42.1%, and PC 2 = 18.5% of total variance). Approximately half of the
HYPOT samples overlapped with the CONT group cluster, while the others spread to the
left or/and downwards, causing a large HYPOT cluster (Figure 6a). This shift is mainly
due to the contribution of fatty acid, lipid, and lipoprotein concentrations, as shown in
the PCA loadings plot (Supplementary Figure S2c). A similar clustering was seen in the
PLS-DA scores plot (Figure 6b). In addition to phenylalanine and histidine, within the
top 20 parameters identified by PLS-DA VIP scores of the first component, fatty acids,
cholesterol subtypes, and HDL fractions were most relevant for separation (Figure 6c).
In the hierarchical cluster analysis, one large cluster consisting mainly of CONT
samples formed, while samples of the HYPOU group were mainly assigned to another
cluster. One smaller but very distinct cluster consisted of samples from all three groups.
The samples from the HYPOT group were dispersed among all clusters (Figure 6d).
Machine learning methods could correctly classify 63–78% of the samples based solely
on the serum metabolomics data from the respective groups (Supplementary Table S2).
To that end, 78% of the samples were assigned to the correct group in the simple logistic
regression model (Table 2c).
The increased serum phenylalanine and histidine concentrations in the group with
hypoadrenocorticism decreased with treatment (Figure 4a,c). The slightly increased acetate
and citrate concentrations seen in the HYPOU group tended to decrease with treatment,
albeit changes were not significant. The slightly increased serum GlycA concentrations
in HYPOU dogs further rose during treatment; also here the increase was not significant
(Figure 4d, Supplementary Table S1).
The slightly lower total serum cholesterol concentrations in the HYPOU group in-
creased in the HYPOT group, whereas the increase in triglyceride concentrations in the
HYPOT group was not found significant. Decreased concentrations of small HDL, large
HDL, and extra-large HDL particles and their associated lipid fractions in the HYPOU
group rose in the HYPOT group (except small HDL triglyceride concentrations). While
most lipid-associated parameters were within the reference intervals in the HYPOT group,
various LDL- and VLDL-associated lipids and triglyceride subtypes had the upper per-
centile slightly above the reference intervals, even if their increase was not significant
compared to the HYPOU group (Figure 4g,h, Supplementary Table S1).
Likewise, the serum concentrations of total and most individual fatty acids increased
with treatment of hypoadrenocorticism (Figure 4i), except for docosahexaenoic acid, which
was not altered in any group, as well as concentrations of arachidonic and docosapentaenoic
acid. Changes in relative fatty acid concentrations in the HYPOT group compared to
the HYPOU group were minor and included increased relative concentrations of linoleic
acid in the HYPOT group and decreased relative concentrations of docosahexaenoic acid
(Supplementary Table S1).
Metabolites 2022, 12, x FOR PEER REVIEW 11 of 22 
 
tate and citrate concentrations seen in the HYPOU group tended to decrease with treat-
ment, albeit changes were not significant. The slightly increased serum GlycA concentra-
Metabolites 2022, 12, 339 11 of 21
tions in HYPOU dogs further rose during treatment; also here the increase was not signif-
icant (Figure 4d, Supplementary Table S1).  
 
FFiigguurree 66.. MMeteatbaobloolmomic iacnaanlyaslyess eosf soefrusemru smamspalmesp ilne sthien gtrhoeupgsro ouf phsypoof ahdyrpenooadcorertnicoicsomrt uicnistrmeautend- 
(HYPOU, n = 35), hyperadrenocorticism treated (HYPOT, n = 23), and the control group (CONT, n = 
treated (HYPO , n = 35), hyperadrenocorticism treated (HYPO , n = 23), and the control group
40). (a) Scores pUlots of principal component analysis (PCA) and (b)T partial least squares-discriminant 
(aCnOalNysTi,sn (P=L4S0-)D. (Aa)) Sbcaosreeds opnlo tms eotfapbroilnocmipiacls cdoamtap oonf esnetruanma lsyasmisp(PleCs Afr)oamnd d(obg)sp ianr tHialYlPeaOst squU (greaernes),- 
dHisYcPriOmTi (nraendt),a annadly tshise (CPOLSN-DT Agr)obuapses d(bolunem). eSthaabdoleodm ciicrscldesa traeporfesseernutm 95s%am cponlefsidfreonmced inotgesrvinalHs,Y wPhOilUe 
(cgorleoerned), HdoYtPs OillTu(srterdat)e,  aindivthideuCaOl sNaTmgprloesu.p Tsh(eb lauxee)s.  Sahrea dlaebdelceirdc lbeys rtehpe rfeisrestn at n9d5% seccoonnfidd (epnrcineciinptaelr)- 
vcoalms,pwonheilnetcs owloitrhe dthdeo ptsericlleunsttargatees ionfd viavriidaunacles oafm thpele dsa. tTah eexapxlaeisnaerde blayb tehlaetd cboymtphoenfiernstt iann pdasrencothned-
(spersi.n (cci)p Vaal)ricaobmlep iomnpeonrttsawnciteh inth perpoejerccteionnta (gVeIsPo) fsvcoarreiasn ocfe cofmthpeondeantat 1e xopf ltahine ePdLbSy-DthAa itdceonmtipfioesn ethnet  
itnopp a2r0e ndtihscersiems.in(act)inVga rpiaabrlaemimeteprosr tinan dceesicnenpdroinjegc toiordne(rV oIfP )imscpoorretsanocfec.o Tmhpe ocnoelonrte1do lfegtheendP LoSn- DthAe 
irdigehntt iifinedsictahteest othpe2 r0eldaitsivcrei mabiunnatdinangcpe aorfa vmaertiearbsleisn, wdeitshc erneddi nangdo brdlueer ionfdiimcaptionrgta hnicgeh.  aTnhde lcoowlo vreadl-
lueegse,n rdesopnectthiveerliyg,h wt ihnidlei cbaetiegset ihlleursetrlaattievse naebuutrnadl avnacleueosf. v(da)r iDabelneds,rowgirtahmre odf ahniedrabrlcuheicinadl cicluatsitnegr ahnigahl-
ysis of serum metabolomic results from canine samples in either HYPOU (green), HYPOT (red), or 
and low values, respectively, while beige illustrates neutral values. (d) Dendrogram of hierarchical
the CONT group (blue). Each number on the x-axis reflects one serum sample. The y-axis shows the 
csliumsitlearraitnya lleyvseislso efxsperruesmsemd eatsa bEoulcolmidiecarne sduilsttsanfrcoems. cHaonrinizeosnatmalp alnesd ivneerittihcaelr lHinYePs OdeUp(igcrt ecelnu)s,teHrYinPgO oTf 
(sraemd)p, loers tahnedC dOifNfeTregnrcoeus pin( bthluee d).isEtaanchcens,u rmesbpeercotinvethlye. x-axis reflects one serum sample. The y-axis
shows the similarity levels expressed as Euclidean distances. Horizontal and vertical lines depict
clusteTrihneg osflisgahmtlpyl elsoawnedrd tiofftearel nsceersuimn t hcehodlisetsatnecreosl, rceosnpceectnivtrealyti.ons in the HYPOU group in-
3c.reDaissecdu sinsi othne HYPOT group, whereas the increase in triglyceride concentrations in the 
HYPOT group was not found significant. Decreased concentrations of small HDL, large 
HDLH, aynpdo aedxrterna-olcaorgrtei cHismDL, a plsaortriecfleersr eadndto thaseiard arsesnoacliaintesdu ffilicpiiedn cfrya, catniodnhs yipne trhade rHenYoPcOor-U 
tgircoisump, roorsCe uinsh tihneg ’Hs YsyPnOdr ogmroeu,pc a(nexbceepcat ussmedallb yHvDaLr iotruigsldyciseorriddee rcsowncitT ehnitnratthioenpsi)t.u Witahriyle– 
amdorsetn laipl iadx-iassosrocciaanteadr ipsaeriaamtreotgeerns iwcaelrley wbyithmine dthicea rleoferrseunrcgei cinatleirnvtaelrsv ienn tthioen Hs Y[5P–O8]. WhileT group, 
diagnoses of these endocrinopathies are primarily based upon hormonal testing, identifica-
tion of various abnormalities in routine blood tests and imaging results can further delineate
cause, severity, and complications [5–8]. In addition, these tests can be used to clinically
 monitor and adjust therapeutic interventions for hypo- and hyperadrenocorticism [5–8].
Metabolites 2022, 12, 339 12 of 21
It is well recognized that endogenous and exogenous glucocorticoid imbalances can
lead to profound metabolic dysfunction with varied severity of systemic illnesses in humans
and animals [5–8]. While such dysfunction is expected to have a major impact on the
metabolome, there is currently a paucity of data on the impact of excess or deficient
corticosteroid levels on the serum metabolome of humans and dogs with these adrenal
endocrinopathies, either untreated or during treatment [11–16].
Utilizing a validated 1H NMR method for dogs, we assessed serum samples from
untreated and treated dogs with hyper- and hypoadrenocorticism. Our findings provide
the first evidence of often contrasting metabolite abnormalities and distinct metabolomic
patterns in these canine adrenal endocrinopathies and reveal that these metabolic changes
do not completely return to baseline with treatment in either condition. The metabolomic
changes reported here will most probably not replace the current diagnostic tests for canine
adrenal endocrinopathies but suggest that this metabolomic platform has the potential to
further define metabolic dysfunction in hyper- and hypoadrenocorticism and may aid in
clinical monitoring to optimize medical treatment.
Using this novel 1H NMR spectroscopy platform, we documented major abnormalities
in the serum metabolome of dogs with endogenous hyper- and hypoadrenocorticism, with
108 of 123 metabolic measurands differing between the two untreated endocrinopathies
and/or the control group. As cortisol is a powerful catabolic hormone with broad metabolic
effects [2], this number of abnormalities may not be surprising. Moreover, many lipid- and
fatty acid-associated metabolites were increased in serum of dogs with hyperadrenocorti-
cism and decreased with hypoadrenocorticism, highlighting the effects of cortisol on their
metabolism. However, other metabolites were altered similarly in both endocrinopathies,
which could either reflect different metabolic pathways affecting the same metabolites or
general disease-related changes. It should be noted that all observed changes in measur-
ands were modest, with less than three-fold differences from the control group and close to
reference intervals.
Metabolomics data were analyzed by both univariate and multivariate analyses, such
as PCA, PLS-DA, hierarchical cluster analyses, and machine learning methods. The clusters
of the control dogs in PCA and PLS-DA were the tightest, suggesting that the broader
but distinct clusters of samples from dogs with untreated hyper- and hypoadrenocorti-
cism (Figures 1a,b, 5a,b and 6a,b) may be due to different disease stages, duration, or
associated complications, including hypertension, inflammation, cholestatic disease, or
others in dogs with endogenous hyperadrenocorticism [7,21] or hypovolemic shock in
dogs with endogenous hypoadrenocorticism [8,22]. Furthermore, we did not distinguish
between adrenal- and pituitary-dependent hyperadrenocorticism, which may have slightly
different serum metabolomic patterns. In addition, while the great majority of dogs with
endogenous hypoadrenocorticism involve both, a glucocorticoid and mineralocorticoid
deficiency and can show greatly varied clinical manifestation [8,22], we did not attempt
to differentiate in the present study between both, nor between primary and secondary
hypoadrenocorticism, which might also contribute to the broadness of the clusters in PCA
and PLS-DA (Figures 1a,b and 6a,b). Finally, as seen in hierarchical cluster analysis, sam-
ples in the cluster of dogs with hyperadrenocorticism were more distant from the cluster of
samples of dogs with hypoadrenocorticism and from the control group, indicating greater
differences in the serum metabolome of these dogs (Figure 2).
Many metabolic abnormalities in serum lipid-associated parameters, e.g., cholesterol
subtypes, HDL-associated lipids, and absolute fatty acid concentrations in samples of dogs
with hyper- and hypoadrenocorticism were noted (Supplementary Table S1). Changes were
often in the opposite direction, with higher and lower serum concentrations in hyper- and
hypoadrenocorticism, respectively. These measurands markedly influenced the separation
of cluster analyses, as reflected by PLS-DA VIP scores (Figure 1d). Lipid-associated abnor-
malities, including hypercholesterolemia and hypertriglyceridemia occur in humans and
dogs with hyperadrenocorticism and during exogenous glucocorticoid exposure [6,7,23–27],
and are possibly caused by glucocorticoid effects on lipolysis, free fatty acid production,
Metabolites 2022, 12, 339 13 of 21
and VLDL synthesis [24]. Furthermore, glucocorticoids promote cholesterol synthesis
through enzyme induction in rat hepatocytes [28,29]. Dogs with untreated hyperadreno-
corticism also showed increased serum cholesterol and triglyceride concentrations in this
study, albeit the median values were still in the reference intervals. However, in a re-
cent experimental lipidomic study of tetracosactide-induced hypercortisolism in dogs
increased total and specific fractions of plasma triglycerides have been found, but not total
cholesterol concentrations [30].
The VLDL and LDL cholesterol concentrations in the group with hyperadrenocorti-
cism were markedly increased, while the increased HDL cholesterol concentrations were
within the reference interval (Figure 4e,g). Additionally, we observed increases in VLDL
and HDL triglyceride concentrations in the group of untreated hyperadrenocorticism. Prior
observations in dogs with hyperadrenocorticism demonstrated mainly increased VLDL
cholesterol fractions but decreased HDL cholesterol and triglyceride fractions (percent-
age distribution), and also absolute increased concentrations of VLDL cholesterol and
triglycerides [23], and mainly increased LDL cholesterol concentrations [31]. Humans with
Cushing’s syndrome show increased LDL and VLDL concentrations leading to hyperc-
holesterolemia and hypertriglyceridemia [6], while HDL cholesterol concentrations were
not found to be decreased [25].
Low serum cholesterol concentrations have previously been reported in dogs with hy-
poadrenocorticism [22] and based upon this study, are mainly characterized by a decrease in
HDL cholesterol concentrations (Figure 4e, Supplementary Table S1). Suggested causes for
the changes in cholesterol concentrations include decreased lipid absorption by the gastroin-
testinal tract, decreased fatty acid mobilization due to low cortisol, and increased utilization
of fatty acids related to high adrenocorticotropic hormone (ACTH) concentrations [22]. As
glucocorticoids induce enzymes that enhance cholesterol synthesis [28,29], their lack may
contribute to low cholesterol levels. Human patients cured from Cushing’s disease exhibit
persistently altered lipid markers [32,33]. Similarly, we found that while many HDL- and
LDL-associated lipid changes were reversible with treatment of hyperadrenocorticism,
many VLDL-associated lipids remained elevated (Supplementary Table S1).
Dogs with treated hypoadrenocorticism also showed altered lipid and lipoprotein
profiles compared to the untreated dogs. However, concentrations of triglycerides and
predominantly VLDL-associated lipids tended to increase, despite being unaltered prior to
treatment. Although these increases were not statistically significant, the upper percentiles
of these parameters were often above the reference intervals (Figure 4g,h, Supplemen-
tary Table S1). Such finding could be related to the administration of glucocorticoids,
which could lead to the secretion of VLDL and increase in triglyceride concentrations [23],
and administration of fludrocortisone or fludrocortisone and prednisone also increased
cholesterol and triglyceride concentrations without inducing clinical features of iatrogenic
hyperadrenocorticism in treated dogs [9].
This 1H NMR platform determines the total concentration of free (non-esterified) and
esterified fatty acids. The high and low total and specific fatty acid concentrations observed
in hyper- and hypoadrenocorticism (Figure 4i, Supplementary Table S1) depend on lipid
concentrations and also on the lipoprotein profile, as the different lipoproteins and their
lipids are esterified with distinct fatty acids [34]. Furthermore, glucocorticoids may also
increase free fatty acids [24,35] and may lead to further increased fatty acid concentrations
in dogs with hyperadrenocorticism.
The 1H NMR platform utilized in this study only identifies nine amino acids, and the
observed differences from the apparently healthy control group and canine reference inter-
val were rather small (Supplementary Table S1). Nevertheless, the increase in serum alanine
concentrations in canine hyperadrenocorticism in our study (Figure 4b) was similar to that
in human patients with Cushing’s disease and prednisolone administration [36,37]. This
was associated with reduced protein synthesis and insulin resistance, which may stimulate
the glucose–alanine cycle [36]. In contrast, the increased serum BCAA, phenylalanine, and
tyrosine concentrations seen in dogs with hyperadrenocorticism (Supplementary Table S1)
Metabolites 2022, 12, 339 14 of 21
were only partially consistent with metabolic patterns seen in human patients with Cush-
ing’s syndrome or disease [12,13,36] and may be related to species-specific metabolic
differences. Dogs with hypoadrenocorticism had increased serum phenylalanine and
histidine concentrations (Figure 4a,c). The unaltered histidine concentrations in dogs
with hyperadrenocorticism are in contrast to a human study that found decreased serum
histidine concentrations in patients with Cushing’s syndrome [12].
Serum GlycA is an inflammatory biomarker that reflects the signals of N-acetylglucosamine
residues within certain acute-phase proteins, mainly α1-acid glycoprotein, α1-antitrypsin,
α1-antichymotrypsin, haptoglobin, and transferrin [38,39]. While there are no prior studies
on serum GlycA concentrations in dogs with altered cortisol levels, the increased GlycA
concentrations seen in dogs with hyperadrenocorticism in our study (Figure 4d), may be
equated with the previously reported increased haptoglobin concentrations in canine hyper-
adrenocorticism [40–43]. While the GlycA concentrations fell slightly following treatment
of hyperadrenocorticism, they remained increased over controls (Figure 4d). Likewise,
serum haptoglobin concentrations decreased in dogs with treated hyperadrenocorticism,
but were still elevated, potentially due to poor control of hyperadrenocorticism, cortisol
precursors, or secondary effects of hyperadrenocorticism [40,41,44]. In human pediatric pa-
tients with Cushing’s disease, serum GlycA concentrations declined after transsphenoidal
surgery [33]. We also found that serum GlycA concentrations are increased in dogs with
hypoadrenocorticism (Figure 4d); however, the mechanisms responsible are unknown, and
clinical implications remain unclear.
While a prior study predicted canine hyperadrenocorticism based on demographic
data, clinical signs, and liver enzyme activities [45] and another predicted canine hy-
poadrenocorticism based on CBC and serum chemistry screening results [46], this is the
first study to apply machine learning approaches based solely on metabolomics data.
The applied machine learning tools performed well, predicting the correct groups when
applied to the two untreated endocrinopathies and the control group (Tables 2 and S2).
However, hyper- and hypoadrenocorticism show different clinical and laboratory char-
acteristics [7,8,10,21], as well as contrasting metabolic features. As such, future studies
would be needed to compare both endocrinopathies to other clinically similar diseases
to determine if this machine learning approach can correctly predict disease based on
metabolomics data. Furthermore, the etiology of these endocrinopathies should be con-
sidered in future machine learning studies to improve classification. In the simple logistic
regression model some samples from dogs with hypoadrenocorticism were misclassified
as control samples (Table 2a), suggesting relatively minor metabolic changes as also re-
flected by dogs with hypoadrenocorticism often showing mild signs that make a prompt
diagnosis difficult [8,10,22]. Machine learning approaches were slightly less accurate in clas-
sifying treated dogs compared to classifying either untreated endocrinopathy or controls
(Tables 2b,c and S2). A false classification of samples from treated dogs could be due to nor-
malization of metabolic abnormalities during treatment (misclassified as control sample),
or due to insufficient resolution of metabolic changes (misclassified as untreated sample).
Although we expected normalization of metabolic changes in serum of treated dogs,
this first of its kind study revealed a more complex picture of serum metabolomic changes
following treatment of dogs with hyper- and hypoadrenocorticism. While some parameters
regained values within the reference interval comparable to the control group, others
were persistently abnormal. In fact, parameters that tended to be overcompensated in
treated dogs with hypoadrenocorticism (no statistical significance), such as different types
of triglycerides or VLDL-associated metabolites, only showed partial reversal in treated
dogs with hyperadrenocorticism (Figure 4g,h, Supplementary Table S1). As these changes
appear to be consistent with glucocorticoid effects, it suggests that some dogs may not
have been treated long enough or may not have been ideally managed. However, as the
duration between diagnosis and sampling for the treated endocrinopathy groups was
not standardized, this might also affect the changes in the serum metabolome. Short-
and long-term induced canine hypercortisolism also showed some different lipidomic
Metabolites 2022, 12, 339 15 of 21
alterations [30]. While the medical treatment of dogs with endogenous adrenal diseases
is clinically monitored for efficacy and safety by hormonal and routine blood tests in
addition to clinical signs, treatment of both endocrinopathies can be challenging [7,8]. For
example, ACTH-ST results for monitoring post-trilostane cortisol levels during treatment
for hyperadrenocorticism were inconsistent with clinical signs [47,48]. However, in this
study, the treated hyperadrenocorticism group showed relatively low pre- and post-ACTH-
ST cortisol levels, and dogs with post ACTH-ST cortisol of less than 10 ng/mL (1 µg/dL)
may be overtreated [7]. Additionally, due to the use of left-over samples the time between
medical treatment and sampling remains unknown and might be different. In treatment of
canines with hypoadrenocorticism, dosage for mineralocorticoids can be monitored with
electrolyte concentrations, but glucocorticoid dosages are adapted according to clinical
signs, so over- and undertreatment may occur [8,49,50].
In addition to the limitations of our study mentioned above, our comparative inves-
tigation of the serum metabolomics of treated and untreated dogs with either hyper- or
hypoadrenocorticism was undertaken with a relatively small number of subjects, which
were not specifically sex- and age-matched, and no attempt was made to differentiate
between etiology, severity, and duration of the endocrinopathy. Dogs with hyperadreno-
corticism were older than those of the other groups. However, the ranges of ages in each
group were broad and overlapping. In a simultaneously performed study on canine hep-
atopathies, we showed that there were no significant differences in serum concentrations
of the metabolomics data between younger and older adult control dogs (expect citrate
concentrations, p = 0.049) [20]. The serum samples of the CONT group used in this study
were also part of the control group in this recent study on canine hepatopathies [20]. Like-
wise, in this study we did not identify significant differences in the metabolomics data by
univariate analysis with Kruskal–Wallis test adjusted with Bonferroni correction and with
PCA of the metabolomics data between two age-dependent CONT subgroups divided at
the age of 6 yrs (Supplementary Figure S1). To simplify the bioinformatic analyses, we
only showed the combined CONT group as we did for our metabolomic study on canine
hepatopathies [20]. However, future prospective studies should consider specific breed-,
sex-, and age-matching.
Similarly, exact treatment regimens (e.g., glucocorticoid +/− mineralocorticoid treat-
ment in the HYPOU group, or any other supportive care) and clinical responses to treatment
were not available, and thus their impact on metabolomics data could not be ascertained.
Likewise, other factors such as body condition score, diet, and time-interval of treatment
were not included. The duration of treatment especially in hypoadrenocorticism might have
an influence on the serum metabolome, as the dosages at the beginning of the treatment
are often higher and then are tapered off [8,9]. As the groups of dogs with hyperadreno-
corticism were not paired, individual influences on the serum metabolome might have
been neglected. As none of the authors were attending veterinarians to any dogs of this
study and left-over samples were used, occult diseases which were not detected by routine
blood test results in the CONT group cannot be fully excluded. Due to the use of left-over
samples the sampling was not standardized and there might be a varied lag period from
sample collection and separation of serum from clot until chilling and freezing which
could affect certain metabolites results. Also, feeding and fasting were not standardized.
However, the metabolic abnormalities that we observed during treatment potentially iden-
tify metabolomic testing as a novel approach to guide and adjust treatment of adrenal
endocrinopathies more precisely. Finally, it will need to be determined if our findings are
consistent with other metabolomic platforms to further evaluate its promise in assessing
disease and treatment of adrenal diseases.
4. Materials and Methods
4.1. Samples and Groups
The study was conducted using left-over serum samples submitted between May
2020 and June 2021 for routine testing to a veterinary diagnostic laboratory (Laboklin
Metabolites 2022, 12, 339 16 of 21
GmbH & Co. KG, Bad Kissingen, Germany). Untreated and treated adult dogs with
hyper- and hypoadrenocorticism, as well as apparently healthy adult dogs were included
in this study. The use of left-over samples for research purposes was approved by the
government in Lower Franconia, Bavaria, Germany (RUF-55.2.2-2532-1-86-5). Laboklin’s
electronic database was searched for results LDDST and ACTH-ST, supporting an adrenal
disorder [8,51,52]. Only basal serum samples from LDDST or ACTH-ST were analyzed in
metabolomic analyses.
The following groups of dogs were assessed in this study:
• HYPERU group—samples from dogs with high serum cortisol concentrations of
>10 ng/mL (>1 µg/dL) in both pre- and 8 h post-LDDST samples supportive of
a diagnosis of hyperadrenocorticism [51,52].
• HYPERT group—samples from treated dogs with hyperadrenocorticism and serum
cortisol concentrations of <20 ng/mL (<2 µg/dL) in pre- and post-ACTH-ST samples.
All HYPERT dogs were different from the HYPERU dogs (unpaired samples).
• HYPOU group—samples with low serum cortisol concentrations of <10 ng/mL
(<1 µg/dL) in both pre- and post-ACTH-ST samples consistent with a diagnosis
of hypoadrenocorticism [8].
• HYPOT group—samples from dogs in the HYPOU group mentioned above were
examined once during treatment for at least two weeks. For those dogs, routine blood
testing during treatment was offered to attending clinicians (free of charge), and the
left-over serum sample was used for the metabolomic study (paired samples).
• CONT group—samples from adult dogs with serum chemistry panel and CBC results
in the reference intervals. All serum samples from apparently healthy dogs were also
part of the control group in our recent metabolomic study on canine hepatopathies [20].
No metabolomic differences were observed between two age-dependent CONT sub-
groups (dogs < 6 yrs old vs. dogs ≥ 6 yrs old) by univariate testing with Kruskal–Wallis
test adjusted with Bonferroni correction and by PCA of their serum metabolomics data
(Supplementary Figure S1). Thus, to simplify the presentation the control dogs were
combined to one CONT group for bioinformatic analyses of the metabolomics data.
Available information on breed, age, sex, neutering status, and other data received
from submission forms, medical consult service, as well as blood test results were gathered
and reviewed. For both untreated endocrinopathies (HYPERU and HYPOU) groups, only
samples from dogs without known concurrent diseases (e.g., infectious diseases, other
specific organ diseases) and for the CONT group only samples without laboratory evidence
of any disease were included. Furthermore, clinical information from submission forms
and from contacting the submitting veterinary clinicians by Laboklin’s medical consult
service was obtained to support the diagnosis of hyper- or hypoadrenocorticism in dogs
of the HYPERU and HYPOU groups, respectively, and to exclude other diseases and prior
treatment with glucocorticoids or trilostane in the HYPOU group.
The laboratory’s inventory of frozen samples was screened for left-over serum samples
with a residual volume of ≥300 µL. These serum samples were originally submitted to
the laboratory after centrifugation and removal of clotted blood and were delivered either
chilled or unchilled if transport time was ≤1 day. Samples with hemolysis and/or icterus
were excluded. Frozen serum samples were thawed, aliquoted (1.8 mL CryoPure tubes,
Sarstedt AG & Co. KG, Nürnbrecht, Germany), and refrozen at −80 ◦C until shipment
for metabolomic analysis within ≤6 months. Results of serum chemistry analyses (Cobas
8000 c701 analyzer, Roche Diagnostics, Mannheim, Germany) and CBC (ADVIA 2120i,
Siemens Healthcare GmbH, Erlangen, Germany or Sysmex XT2000i, Sysmex Deutschland
GmbH, Norderstedt, Germany) were reviewed. Serum chemistry analyses were performed
on thawed samples, if not already undertaken during routine testing. Cortisol measure-
ments for LDDST and ACTH-ST were conducted with a Cobas 8000 e602 analyzer with an
electrochemiluminescence immunoassay (Roche Diagnostics, Mannheim, Germany).
Metabolites 2022, 12, 339 17 of 21
4.2. Serum Metabolomic Analyses
The metabolomic analysis of canine serum samples was performed as previously
described in [19]. Serum samples were shipped frozen overnight on ice packs to PetMeta
Labs Oy (Helsinki, Finland). Targeted metabolomic analysis was conducted with a 1H NMR
spectrometer (Bruker AVANCE III HD 500 MHz, Bruker Corp., Billerica, MA, USA). The
1H NMR method used is optimized for dogs and validated for canine serum and plasma
samples [19]. A similar 1H NMR method has been described and largely utilized for human
serum and plasma samples [53]. Metabolomics data were reported after spectral processing
as metabolite concentrations, and ratios and percentages were calculated. Unnamed peaks
were not reported and thus not included in further analyses.
4.3. Statistical Analysis
4.3.1. Univariate Analyses
Univariate statistical analyses were performed using MS Office Excel (Microsoft Corp.,
Redmond, WA, USA) and SPSS Statistics (version 26; IBM Corp., Armonk, NY, USA)
software programs. All continuous data were assessed for normal distribution. Differences
in age were evaluated using a one-way analysis of variance (ANOVA) [54]. Differences in
sex and neutering status were evaluated using chi-square tests.
Concentrations of metabolomics data missing at random were imputed by the median
of the corresponding variable, and concentrations below the detection limit were imputed
with a zero value. Differences in metabolomics data were assessed with Kruskal–Wallis
test for comparison of unpaired samples (CONT, HYPERU, HYPERT, and HYPOU groups,
as well as for the age-dependent CONT subgroups, HYPERU, HYPERT, and HYPOU
groups) [55] and with a Wilcoxon signed-rank test for the paired samples of HYPOU and
HYPOT [56], both adjusted with a Bonferroni correction [57]. The level of significance was
set at p < 0.05.
While in the Kruskal–Wallis test and multivariate analyses all collected HYPOU sam-
ples (n = 35) are included, the Wilcoxon signed-rank test includes only the paired HYPOU
and HYPOT samples (n = 23).
4.3.2. Multivariate Analyses
Imputation of serum metabolomics data was completed as described above. PCA [58],
PLS-DA [59], hierarchical cluster analyses, and the hierarchical cluster heatmap of the
serum metabolomics data were performed using MetaboAnalyst 5.0 [60] with auto-scaled
variables. Hierarchical cluster analyses were performed using the Ward clustering algo-
rithm and the Euclidean distance measure [61]. A hierarchical cluster heatmap was created
to visualize changes in the 20 most discriminative parameters identified by VIP scores
in PLS-DA. Machine learning methods were performed with Waikato Environment for
Knowledge Analysis (WEKA) 3.95 [62]. Models applied were simple logistic regression [63],
support vector machines [64], k-nearest neighbors (KNN) algorithm [65], Multilayer Per-
ceptron (MLP) Classifier [66], Random Forest [67], and multinomial naïve Bayes [68]. The
default settings of the parameters for the respective WEKA implementation were used for
all machine learning methods. Machine learning models were evaluated using 10-fold full
cross-validation for each model.
5. Conclusions
Using a targeted metabolomic 1H NMR platform quantifying 123 metabolic parame-
ters, this study revealed distinct metabolomic patterns and major metabolic abnormalities
in the serum of dogs with untreated and treated hyper- or hypoadrenocorticism. Serum
amino acid concentrations changed in various directions, with serum phenylalanine concen-
trations being increased in both endocrinopathies, while serum concentrations of tyrosine,
alanine, and total branched-chain amino acid were only increased in hyperadrenocorticism,
and histidine concentrations were elevated in hypoadrenocorticism. Various lipoprotein
and lipid fractions, and fatty acid concentrations were often opposingly altered and were
Metabolites 2022, 12, 339 18 of 21
predominantly increased in hyperadrenocorticism and decreased in hypoadrenocorticism.
These metabolic changes may give new insights in the pathophysiology and improve char-
acterization of these endocrinopathies. It remains unclear why the metabolic alterations
were only partially reversed following treatment, so further investigations are warranted
to enhance our understanding of disease management. Further optimization of applied
machine learning approaches may facilitate future diagnosis or improve monitoring of
treatment outcomes for these patients.
Supplementary Materials: The following supporting information can be downloaded at: https:
//www.mdpi.com/article/10.3390/metabo12040339/s1, Table S1: Metabolomic serum parameters
significantly differing between dogs in the groups of CONT (n = 40), HYPERU (n = 27), HYPERT
(n = 28), HYPOU (n = 35), and HYPOT (n = 23); Table S2: Different machine learning models classifying
groups based solely on metabolomics data from serum samples of dogs in the groups of CONT
(n = 40), HYPERU (n = 27), and HYPOU (n = 35); HYPERU (n = 27), HYPERT (n = 28), and CONT
(n = 40); HYPOU (n = 35), HYPOT (n = 23), and CONT (n = 40); Table S3: Detailed accuracy by class
for the simple logistic regression model of the metabolomics data from serum samples of dogs in
the groups of CONT (n = 40), HYPERU (n = 27), and HYPOU (n = 35); Figure S1: Comparison of
age and metabolomics data of CONT groups subdivided at the age of 6 years into dogs of younger
(<6 yrs, n = 22) and of older age (≥6 yrs, n = 18); Figure S2: Loadings plot of principal component
analysis based on metabolomics data between serum samples (a) of dogs in the groups of CONT
(n = 40), HYPERU (n = 27), and HYPOU (n = 35); (b) HYPERU (n = 27), HYPERT (n = 28), and CONT
(n = 40); (c) HYPOU (n = 35), HYPOT (n = 23), and CONT (n = 40); Figure S3: Loadings plot of
partial least squares–discriminant analysis (PLS-DA) based on metabolomics data between serum
samples of dogs in the groups of CONT (n = 40), HYPERU (n = 27), and HYPOU (n = 35); Figure S4:
Results of the 10-fold cross-validation of the partial least squares–discriminant analysis (PLS-DA)
model based on metabolomics data between serum samples of dogs in the groups of CONT (n = 40),
HYPERU (n = 27), and HYPOU (n = 35) with R2, Q2, and accuracy measures based on the number
of components; Figure S5: Results of a permutation test with 2000 permutations for the partial least
squares–discriminant analysis (PLS-DA) based on metabolomics data between serum samples of
dogs in the groups of CONT (n = 40), HYPERU (n = 27), and HYPOU (n = 35); Equation S1: Equation
of simple logistic regression model of the metabolomics data from serum samples of dogs in the
groups of CONT (n = 40), HYPERU (n = 27), and HYPOU (n = 35).
Author Contributions: Conceptualization, U.G., C.A.I., C.O., C.W., E.M. and H.L.; formal analysis,
C.A.I., F.D. and G.S.; investigation, C.A.I. and U.G.; resources, H.L., C.W. and E.M.; data curation,
C.A.I. and C.O.; writing—original draft preparation, C.A.I. and U.G.; writing—review and edit-
ing, F.D., C.O., G.S., H.L., C.W. and E.M.; visualization, C.A.I., F.D. and G.S.; supervision, U.G.;
funding acquisition, E.M. and H.L. All authors have read and agreed to the published version of
the manuscript.
Funding: This research received no external funding. Laboklin GmbH & Co. KG, Bad Kissingen,
Germany, provided workspace and materials, enabled collection of left-over blood samples and
conduction of routine blood test analyses. Metabolomic analyses were funded by PetMeta Labs Oy,
Helsinki, Finland. Bioinformatic analyses were performed independent of Laboklin and PetMeta and
were supported by intramural funds within the Institute for Chemistry and Bioanalytics, School of
Life Sciences, University of Applied Sciences Northwestern Switzerland, Muttenz, Switzerland.
Institutional Review Board Statement: The use of left-over blood samples for research purposes is
stated in the general terms and conditions of Laboklin GmbH & Co. KG and was approved by the
Government in Lower Franconia, Bavaria, Germany (RUF-55.2.2-2532-1-86-5).
Informed Consent Statement: Not applicable.
Data Availability Statement: Data is contained within the article or supplementary material. Any
further data may be requested from the corresponding author. The data are not publicly available
due to it contains patient information.
Acknowledgments: We thank the veterinarians who submitted samples to Laboklin GmbH & Co.
KG, thereby enabling the use of left-over blood samples for scientific purposes. We also acknowledge
the regional Laboklin teams, particularly in Germany and Czech Republic, for their assistance in
Metabolites 2022, 12, 339 19 of 21
obtaining medical information. Furthermore, we appreciate scientific editing by Leslie King. Finally,
we would like to dedicate this study and manuscript to Claudia Reusch on her retirement, and in
honor of her seminal contributions, leadership in veterinary endocrinology, and establishment of a
superb small animal specialty clinic at the University of Zürich.
Conflicts of Interest: This study was conducted as part of C.A.I.’s doctoral thesis at the University of
Zürich, Switzerland. C.A.I. and C.W. are employed, and E.M. is the owner and executive director
of Laboklin GmbH & Co. KG, Bad Kissingen, Germany. Laboklin offers a variety of diagnostic
laboratory tests for animals. C.O. is employed, and H.L. is the board director of PetMeta Labs Oy,
Helsinki, Finland, a company offering commercial metabolomic testing for dogs. U.G. has been
a scientific advisor to Laboklin. F.D. is the CEO of Dieterle Life Sciences Consulting, Binningen,
Switzerland. G.S. declares no potential conflicts of interest.
References
1. Kemppainen, R.J.; Behrend, E. Adrenal physiology. Vet. Clin. North Am. Small Anim. Pract. 1997, 27, 173–186. [CrossRef]
2. Hall, J.E.; Hall, M.E. Guyton and Hall Textbook of Medical Physiology, 14th ed.; Elsevier Saunders: Philadelphia, PA, USA, 2021;
ISBN 978-0-323-59712-8.
3. Greco, D.S. Hypoadrenocorticism in small animals. Clin. Tech. Small Anim. Pract. 2007, 22, 32–35. [CrossRef] [PubMed]
4. Merchant, S.R.; Taboada, J. Endocrinopathies: Thyroid and adrenal disorders. Vet. Clin. North Am. Small Anim. Pract. 1997,
27, 1285–1303. [CrossRef]
5. Charmandari, E.; Nicolaides, N.C.; Chrousos, G.P. Adrenal insufficiency. Lancet 2014, 383, 2152–2167. [CrossRef]
6. Arnaldi, G.; Angeli, A.; Atkinson, A.B.; Bertagna, X.; Cavagnini, F.; Chrousos, G.P.; Fava, G.A.; Findling, J.W.; Gaillard, R.C.;
Grossman, A.B.; et al. Diagnosis and complications of Cushing’s syndrome: A consensus statement. J. Clin. Endocrinol. Metab.
2003, 88, 5593–5602. [CrossRef]
7. Behrend, E.N. Canine Hyperadrenocorticism. In Canine and Feline Endocrinology, 4th ed.; Feldman, E.C., Nelson, R.W., Reusch, C.,
Scott-Moncrieff, J.C., Eds.; Elsevier: St. Louis, MO, USA, 2015; pp. 377–451. ISBN 9781455744565.
8. Scott-Moncrieff, J.C. Hypoadrenocorticism. In Canine and Feline Endocrinology, 4th ed.; Feldman, E.C., Nelson, R.W., Reusch, C.,
Scott-Moncrieff, J.C., Eds.; Elsevier: St. Louis, MO, USA, 2015; pp. 485–520. ISBN 9781455744565.
9. Melián, C.; Peterson, M.E. Diagnosis and treatment of naturally occurring hypoadrenocorticism in 42 dogs. J. Small Anim. Pract.
1996, 37, 268–275. [CrossRef]
10. Peterson, M.E.; Kintzer, P.P.; Kass, P.H. Pretreatment clinical and laboratory findings in dogs with hypoadrenocorticism: 225 cases
(1979–1993). J. Am. Vet. Med. Assoc. 1996, 208, 85–91.
11. Muñoz-Prieto, A.; Rubić, I.; Horvatic, A.; Rafaj, R.B.; Cerón, J.J.; Tvarijonaviciute, A.; Mrljak, V. Metabolic profiling of serum from
dogs with pituitary-dependent hyperadrenocorticism. Res. Vet. Sci. 2021, 138, 161–166. [CrossRef]
12. Di Dalmazi, G.; Quinkler, M.; Deutschbein, T.; Prehn, C.; Rayes, N.; Kroiss, M.; Berr, C.M.; Stalla, G.; Fassnacht, M.;
Adamski, J.; et al. Cortisol-related metabolic alterations assessed by mass spectrometry assay in patients with Cushing’s
syndrome. Eur. J. Endocrinol. 2017, 177, 227–237. [CrossRef]
13. Vega-Beyhart, A.; Iruarrizaga, M.; Pané, A.; García-Eguren, G.; Giró, O.; Boswell, L.; Aranda, G.; Flores, V.; Casals, G.;
Alonso, C.; et al. Endogenous cortisol excess confers a unique lipid signature and metabolic network. J. Mol. Med. 2021,
99, 1085–1099. [CrossRef]
14. Chantzichristos, D.; Svensson, P.-A.; Garner, T.; Glad, C.A.; Walker, B.R.; Bergthorsdottir, R.; Ragnarsson, O.; Trimpou, P.;
Stimson, R.H.; Borresen, S.W.; et al. Identification of human glucocorticoid response markers using integrated multi-omic
analysis from a randomized crossover trial. Elife 2021, 10, e62236. [CrossRef]
15. Nguyen, L.S.; Prifti, E.; Ichou, F.; Leban, M.; Funck-Brentano, C.; Touraine, P.; Salem, J.-E.; Bachelot, A. Effect of congenital adrenal
hyperplasia treated by glucocorticoids on plasma metabolome: A machine-learning-based analysis. Sci. Rep. 2020, 10, 8859.
[CrossRef] [PubMed]
16. Alwashih, M.A.; Watson, D.G.; Andrew, R.; Stimson, R.H.; Alossaimi, M.; Blackburn, G.; Walker, B.R. Plasma metabolomic profile
varies with glucocorticoid dose in patients with congenital adrenal hyperplasia. Sci. Rep. 2017, 7, 17092. [CrossRef] [PubMed]
17. Hollywood, K.; Brison, D.R.; Goodacre, R. Metabolomics: Current technologies and future trends. Proteomics 2006, 6, 4716–4723.
[CrossRef] [PubMed]
18. Dunn, W.B.; Broadhurst, D.I.; Atherton, H.J.; Goodacre, R.; Griffin, J.L. Systems level studies of mammalian metabolomes: The
roles of mass spectrometry and nuclear magnetic resonance spectroscopy. Chem. Soc. Rev. 2011, 40, 387–426. [CrossRef]
19. Ottka, C.; Vapalahti, K.; Puurunen, J.; Vahtera, L.; Lohi, H. A novel canine nuclear magnetic resonance spectroscopy-based
metabolomics platform: Validation and sample handling. Vet. Clin. Pathol. 2021, 50, 410–426. [CrossRef]
20. Imbery, C.A.; Dieterle, F.; Ottka, C.; Weber, C.; Schlotterbeck, G.; Müller, E.; Lohi, H.; Giger, U. Metabolomic serum abnormalities
in dogs with hepatopathies. Sci. Rep. 2022, 12, 5329. [CrossRef]
21. Bennaim, M.; Shiel, R.E.; Mooney, C.T. Diagnosis of spontaneous hyperadrenocorticism in dogs. Part 1: Pathophysiology,
aetiology, clinical and clinicopathological features. Vet. J. 2019, 252, 105342. [CrossRef]
Metabolites 2022, 12, 339 20 of 21
22. Thompson, A.L.; Scott-Moncrieff, J.C.; Anderson, J.D. Comparison of classic hypoadrenocorticism with glucocorticoid-deficient
hypoadrenocorticism in dogs: 46 cases (1985–2005). J. Am. Vet. Med. Assoc. 2007, 230, 1190–1194. [CrossRef]
23. Jericó, M.M.; de Camargo Chiquito, F.; Kajihara, K.; Moreira, M.A.B.; Gonzales, R.; Machado, F.L.A.; Nunes, V.S.; Catanozi, S.;
Nakandakare, E.R. Chromatographic analysis of lipid fractions in healthy dogs and dogs with obesity or hyperadrenocorticism.
J. Vet. Diagn. Investig. 2009, 21, 203–207. [CrossRef]
24. Arnaldi, G.; Scandali, V.M.; Trementino, L.; Cardinaletti, M.; Appolloni, G.; Boscaro, M. Pathophysiology of dyslipidemia in
Cushing’s syndrome. Neuroendocrinology 2010, 92, 86–90. [CrossRef]
25. Mancini, T.; Kola, B.; Mantero, F.; Boscaro, M.; Arnaldi, G. High cardiovascular risk in patients with Cushing’s syndrome
according to 1999 WHO/ISH guidelines. Clin. Endocrinol. 2004, 61, 768–777. [CrossRef] [PubMed]
26. Zimmerman, J.; Fainaru, M.; Eisenberg, S. The effects of prednisone therapy on plasma lipoproteins and apolipoproteins: A
prospective study. Metabolism 1984, 33, 521–526. [CrossRef]
27. Tinklenberg, R.L.; Murphy, S.D.; Mochel, J.P.; Seo, Y.-J.; Mahaffey, A.L.; Yan, Y.; Ward, J.L. Evaluation of dose-response effects of
short-term oral prednisone administration on clinicopathologic and hemodynamic variables in healthy dogs. Am. J. Vet. Res.
2020, 81, 317–325. [CrossRef] [PubMed]
28. Wiss, O.; Wiss, V. The stimulation of the cholesterol synthesis in rat liver by hydrocortisone. Helv. Chim. Acta 1976, 59, 2179–2180.
[CrossRef]
29. Lin, R.C.; Snodgrass, P.J. Effect of dexamethasone on 3-hydroxy-3-methylglutaryl-coenzyme a reductase activity and cholesterol
synthesis in rat liver. Biochim. Biophys. Acta 1982, 713, 240–250. [CrossRef]
30. Sieber-Ruckstuhl, N.S.; Burla, B.; Spoerel, S.; Schmid, F.; Venzin, C.; Cazenave-Gassiot, A.; Bendt, A.K.; Torta, F.; Wenk, M.R.;
Boretti, F.S. Changes in the canine plasma lipidome after short- and long-term excess glucocorticoid exposure. Sci. Rep. 2019,
9, 6015. [CrossRef]
31. Barrie, J.; Watson, T.D.G.; Stear, M.J.; Nash, A.S. Plasma cholesterol and lipoprotein concentrations in the dog: The effects of age,
breed, gender and endocrine disease. J. Small Anim. Pract. 1993, 34, 507–512. [CrossRef]
32. Colao, A.; Pivonello, R.; Spiezia, S.; Faggiano, A.; Ferone, D.; Filippella, M.; Marzullo, P.; Cerbone, G.; Siciliani, M.; Lombardi, G.
Persistence of increased cardiovascular risk in patients with Cushing’s disease after five years of successful cure. J. Clin. Endocrinol.
Metab. 1999, 84, 2664–2672. [CrossRef]
33. Makri, A.; Cheung, A.; Sinaii, N.; Remaley, A.T.; Sampson, M.; Keil, M.; Belyavskaya, E.; Lyssikatos, C.; de La Luz Sierra, M.;
Stratakis, C.A.; et al. Lipoprotein particles in patients with pediatric Cushing disease and possible cardiovascular risks. Pediatr.
Res. 2019, 86, 375–381. [CrossRef]
34. Maldonado, E.N.; Romero, J.R.; Ochoa, B.; Aveldaño, M.I. Lipid and fatty acid composition of canine lipoproteins. Comp. Biochem.
Physiol. B. Biochem. Mol. Biol. 2001, 128, 719–729. [CrossRef]
35. Macfarlane, D.P.; Forbes, S.; Walker, B.R. Glucocorticoids and fatty acid metabolism in humans: Fuelling fat redistribution in the
metabolic syndrome. J. Endocrinol. 2008, 197, 189–204. [CrossRef] [PubMed]
36. Faggiano, A.; Pivonello, R.; Melis, D.; Alfieri, R.; Filippella, M.; Spagnuolo, G.; Salvatore, F.; Lombardi, G.; Colao, A. Evaluation
of circulating levels and renal clearance of natural amino acids in patients with Cushing’s disease. J. Endocrinol. Investig. 2002,
25, 142–151. [CrossRef] [PubMed]
37. Ellero-Simatos, S.; Szymańska, E.; Rullmann, T.; Dokter, W.H.; Ramaker, R.; Berger, R.; van Iersel, T.M.; Smilde, A.K.;
Hankemeier, T.; Alkema, W. Assessing the metabolic effects of prednisolone in healthy volunteers using urine metabolic profiling.
Genome Med. 2012, 4, 94. [CrossRef] [PubMed]
38. Otvos, J.D.; Shalaurova, I.; Wolak-Dinsmore, J.; Connelly, M.A.; Mackey, R.H.; Stein, J.H.; Tracy, R.P. GlycA: A composite nuclear
magnetic resonance biomarker of systemic inflammation. Clin. Chem. 2015, 61, 714–723. [CrossRef] [PubMed]
39. Bell, J.D.; Brown, J.; Nicholson, J.K.; Sadler, P.J. Assignment of resonances for ‘acute-phase’ glycoproteins in high resolution
proton NMR spectra of human blood plasma. FEBS Lett. 1987, 215, 311–315. [CrossRef]
40. McGrotty, Y.L.; Arteaga, A.; Knottenbelt, C.M.; Ramsey, I.K.; Eckersall, P.D. Haptoglobin concentrations in dogs undergoing
trilostane treatment for hyperadrenocorticism. Vet. Clin. Pathol. 2005, 34, 255–258. [CrossRef]
41. Arteaga, A.; Dhand, N.K.; McCann, T.; Knottenbelt, C.M.; Tebb, A.J.; Evans, H.; Eckersall, P.D.; Ramsey, I.K. Monitoring the
response of canine hyperadrenocorticism to trilostane treatment by assessment of acute phase protein concentrations. J. Small
Anim. Pract. 2010, 51, 204–209. [CrossRef]
42. McGrotty, Y.L.; Knottenbelt, C.M.; Ramsey, I.K.; Reid, S.W.J.; Eckersall, P.D. Haptoglobin concentrations in a canine hospital
population. Vet. Rec. 2003, 152, 562–564. [CrossRef]
43. Caldin, M.; Tasca, S.; Carli, E.; Bianchini, S.; Furlanello, T.; Martinez-Subiela, S.; Cerón, J.J. Serum acute phase protein concentra-
tions in dogs with hyperadrenocorticism with and without concurrent inflammatory conditions. Vet. Clin. Pathol. 2009, 38, 63–68.
[CrossRef]
44. Golinelli, S.; de Marco, V.; Leal, R.O.; Barbarossa, A.; Aniballi, C.; Maietti, E.; Tardo, A.M.; Galac, S.; Fracassi, F. Comparison
of methods to monitor dogs with hypercortisolism treated with trilostane. J. Vet. Intern. Med. 2021, 35, 2616–2627. [CrossRef]
[PubMed]
45. Schofield, I.; Brodbelt, D.C.; Kennedy, N.; Niessen, S.J.M.; Church, D.B.; Geddes, R.F.; O’Neill, D.G. Machine-learning based
prediction of Cushing’s syndrome in dogs attending UK primary-care veterinary practice. Sci. Rep. 2021, 11, 9035. [CrossRef]
[PubMed]
Metabolites 2022, 12, 339 21 of 21
46. Reagan, K.L.; Reagan, B.A.; Gilor, C. Machine learning algorithm as a diagnostic tool for hypoadrenocorticism in dogs.
Domest. Anim. Endocrinol. 2020, 72, 106396. [CrossRef] [PubMed]
47. Wehner, A.; Glöckner, S.; Weiss, B.; Ballhausen, D.; Stockhaus, C.; Zablotski, Y.; Hartmann, K. Association between ACTH
stimulation test results and clinical signs in dogs with hyperadrenocorticism treated with trilostane. Vet. J. 2021, 276, 105740.
[CrossRef]
48. Boretti, F.S.; Holzthüm, J.; Reusch, C.E.; Sieber-Ruckstuhl, N.S. Fehlender Zusammenhang zwischen klinischen Symptomen und
Laborwerten von Hunden mit Hyperadrenokortizismus vor und während der Trilostan-Behandlung. Schweiz. Arch. Tierheilkd.
2016, 158, 631–638. [CrossRef]
49. Kintzer, P.P.; Peterson, M.E. Treatment and long-term follow-up of 205 dogs with hypoadrenocorticism. J. Vet. Intern. Med. 1997,
11, 43–49. [CrossRef]
50. Zeugswetter, F.K.; Haninger, T. Prednisolondosierung bei Hunden mit Hypoadrenokortizismus nach Integration der ACTH-
Messung in die Therapieüberwachung. Tierarztl. Prax. Ausg. K Kleintiere Heimtiere 2018, 46, 90–96. [CrossRef]
51. Behrend, E.N.; Kemppainen, R.J. Diagnosis of canine hyperadrenocorticism. Vet. Clin. North Am. Small Anim. Pract. 2001,
31, 985–1003. [CrossRef]
52. Peterson, M.E. Diagnosis of hyperadrenocorticism in dogs. Clin. Tech. Small Anim. Pract. 2007, 22, 2–11. [CrossRef]
53. Soininen, P.; Kangas, A.J.; Würtz, P.; Tukiainen, T.; Tynkkynen, T.; Laatikainen, R.; Järvelin, M.-R.; Kähönen, M.; Lehtimäki, T.;
Viikari, J.; et al. High-throughput serum NMR metabonomics for cost-effective holistic studies on systemic metabolism. Analyst
2009, 134, 1781–1785. [CrossRef]
54. Doncaster, C.P.; Davey, A.J.H. Analysis of Variance and Covariance; Cambridge University Press: Cambridge, UK, 2007;
ISBN 9780511611377.
55. Kruskal, W.H.; Wallis, W.A. Use of ranks in one-criterion variance analysis. J. Am. Stat. Assoc. 1952, 47, 583–621. [CrossRef]
56. Wilcoxon, F. Individual comparisons by ranking methods. Biom. Bull. 1945, 1, 80–83. [CrossRef]
57. Dunn, O.J. Multiple comparisons among means. J. Am. Stat. Assoc. 1961, 56, 52–64. [CrossRef]
58. Hotelling, H. Analysis of a complex of statistical variables into principal components. J. Educ. Psychol. 1933, 24, 417–441.
[CrossRef]
59. Squillacciotti, S. Prediction oriented classification in PLS path modeling. In Handbook of Partial Least Squares; Esposito Vinzi, V.,
Chin, W.W., Henseler, J., Wang, H., Eds.; Springer: Berlin/Heidelberg, Germany, 2010; pp. 219–233. ISBN 978-3-540-32825-4.
60. MetaboAnalyst. Available online: https://www.metaboanalyst.ca/ (accessed on 6 October 2021).
61. Rokach, L.; Maimon, O. Clustering methods. In Data Mining and Knowledge Discovery Handbook; Maimon, O.Z., Rokach, L., Eds.;
Springer: New York, NY, USA, 2005; pp. 321–352. ISBN 978-0-387-25465-4.
62. Witten, I.H.; Frank, E.; Hall, M.A.; Pal, C.J. Data Mining: Practical Machine Learning Tools and Techniques; The WEKA Workbench.
Online Appendix, 4th ed.; Elsevier Reference Monographs: Cambridge, MA, USA, 2011; ISBN 978-0128042915.
63. Sumner, M.; Frank, E.; Hall, M. Speeding up logistic model tree induction. In Knowledge Discovery in Databases: PKDD 2005: 9th Eu-
ropean Conference on Principles and Practice of Knowledge Discovery in Databases; Hutchison, D., Kanade, T., Kittler, J., Kleinberg, J.M.,
Mattern, F., Mitchell, J.C., Naor, M., Nierstrasz, O., Pandu Rangan, C., Steffen, B., et al., Eds.; Springer: Berlin/Heidelberg,
Germany, 2005; pp. 675–683. ISBN 978-3-540-29244-9.
64. Crammer, K.; Singer, Y. On the algorithmic implementation of multiclass kernel-based vector machines. J. Mach. Learn. Res.
2002, 2, 265–292.
65. Aha, D.W.; Kibler, D.; Albert, M.K. Instance-based learning algorithms. Mach. Learn. 1991, 6, 37–66. [CrossRef]
66. Morariu, D.; Cret,ulescu, R.; Breazu, M. The WEKA multilayer perceptron classifier. J. Adv. Stat. IT&C Econom. Life Sci. 2017, 7.
67. Breiman, L. Random Forests. Mach. Learn. 2001, 45, 5–32. [CrossRef]
68. Hand, D.J.; Yu, K. Idiot’s bayes: Not so stupid after all? Int. Stat. Rev. 2001, 69, 385–398. [CrossRef]