My path

"Shortening the cycle", stated honestly

Where the 12–15 years go

The recurrent breeding cycle gets shorter. The release pipeline mostly keeps its years — what changes is time-to-decision.

The map for this talk

The Potato Genetic Improvement project’s objectives → where I have already done each one

Why a potato audience should care

Heritabilities and correlations

High h² for berry roundness (LvW) and TAcy = easy selection targets; modest total-yield h² = limited room

Upright berry weight tracks plot berry weight; TAcy vs. fruit-rot trade-off

What counts as a “stable” QTL?

Consensus, not p-values, is what a breeder can act on.

Results

Meta-QTL as a community deliverable

Flavonoid accumulation across Vaccinium fruit development (Albert et al. 2023, Fig. 1C)

• Cranberry anthocyanin, proanthocyanidin and color QTL from three populations, synthesized onto one map and anchored to the genome

• A stable chromosome 3 hotspot — where a MYBA-like cluster sits — handed to the Vaccinium flavonoid community

• Delivered through VacciniumCAP: the multi-state blueberry and cranberry fruit-quality project

My role: cross-population QTL synthesis and anchoring · co-author

Albert, Iorizzo, Mengist, Montanari, Zalapa, Maule, Edger, Yocca, Platts, Pucker & Espley 2023 · Plant Physiology 192:1696–1710 · doi.org/10.1093/plphys/kiad250 · VacciniumCAP, USDA-NIFA 2019-51181-30015

From QTL to a community genotyping platform

Same populations, same pipeline: GBS (left) vs Flex-Seq (right) · fruit rot (top), epicuticular wax (bottom)

• A targeted, transferable genotyping panel for cranberry: 17,502 loci · 99.8% recovery · haplotypes, not just SNPs

• Stable meta-QTL from my study went in as design targets

• Fruit-rot resistance: 1 → 4 QTL once the markers transferred

• Parent–offspring checks caught mislabelled parents — germplasm identity in a clonal crop

My role: formal analysis · methodology · software · validation

Clare, S. J., …, Maule, A. F., Zalapa, J., …, Bassil, N. V. 2026 · The Plant Genome 19:e70153 · doi.org/10.1002/tpg2.70153 · USDA-ARS Corvallis · WSU · ARS Chatsworth · Rutgers · MSU · NCSU · UW–Madison

Experimental parameters

• PopulationsCNJ02, CNJ04, GRYG

• Sites2 (Necedah, Tomah WI)

• Flights8 dates, 2018–2019

• Sessionsmultiple per date (repeated measures)

• Ground truth~10% of plots per date: A535 anthocyanin, chlorophyll

• Sensorconsumer RGB (DJI P4P)

Pipeline: flights → orthomosaic → plots → 23 indices

Automated plot identification · semantic segmentation · 23 vegetation indices per plot per session · reproducible (Docker / HTCondor)

The same toolkit, post-harvest · Loarca et al. 2024 · Plant Methods 20:172 · a Breeding Insight collaboration

BerryPortraits

• Open-source, YOLOv8-based segmentation of post-harvest berry images → per-berry color, size, shape and uniformity

• Berry segmentation precision and recall ≥ 0.99; agrees with the prior validated tool at r > 0.94

• Color in CIELAB, so pigment is not confounded with lightness

• Built with the Breeding Insight science and development teams; code public

My role: conceptualization · design · analysis · software testing

Model comparison: random forest wins on accuracy and stability

50 resamples of 80:20 splits

CV of predictions across repeated views of the same plot

From a time series to a selectable index: the LMI

Visual validation: top vs bottom LMI genotypes through the season

GRYG · P40 (high LMI) vs P150 (low LMI) · each column is a flight date

LMI segregates in CNJ02 · genomic h² = 0.74

LMI distributions and Q–Q plots

CNJ02 BLUPs

11 LMI QTL on LG 6–12, each 1.8–9.1% of genetic variance

Mapped on GBLUPs, so effect sizes are conservatively shrunk · polygenic architecture

Candidate genes: the dormancy regulators you would expect

• ClockLHY, PRR95

• PhotoperiodCRY1, COL12

• Dormancy MADSSVP, AGL24

• FloweringSPL15, FTIP1/3

---

Potato translation: tuberization and vine maturity run on the same machinery — StCDF1 (chr 5, released by the clock’s GI–FKF1 module), StCOL, StSP5G → StSP6A, the FT-family tuberigen. A canopy-timing QTL in potato would be expected to land on these families — and gets tested against chr 5 first.

Where it fits

Aim 1 · markers · Objective 3

Markers before the field — a service the breeder and pathologist feel in year one

• Screen seedling families before they take up a hill: Ry_adg / Ry_sto (PVY), the chr 9 TRV locus, GpaIV / Gpa5 / H1 (PCN), PLRV, RMc1 (Columbia root-knot) — already 80% of families carry a Ry parent

• One outsourced assay instead of many: the potato DArTag panel carries the trait markers and a 2.5–4K genome-wide backbone → identity and pedigree QC, R-gene dosage, and genomic-prediction training data from the same run

• Dosage-aware parent choice: a duplex Ry parent gives ~5/6 carriers vs 1/2 from simplex — more resistant seedlings per cross before anyone scores a plant

• Genomic prediction for the polygenic traits — specific gravity, fry color, yield, G. pallida partial resistance — trained on Tri-State multi-site data

SGPG FY2024–25 reports (41 of 163 year-2 entries Ry-confirmed vs 7.5% in 2020; multiplex PCR assay) · Endelman et al. 2024, Plant Genome 17:e20484 (DArTag, polyBreedR) · Anglin et al. 2024, Am. J. Potato Res. (TRV chr 9) · Silvestre et al. 2026 (PCN h² 0.38–0.56)

Aim 2 · measurement · Objectives 1.C and 4

Phenotypes that are numbers — the canopy and the cut tuber, where the markers don’t exist yet

• Canopy time series over early-generation plots and the Aberdeen Tri-State trial: emergence, row closure and senescence → LMI-style onset / rate / area indices, with StCDF1 maturity fitted as a covariate

• Under contrasting N and irrigation → canopy-duration response per clone: the first heritable, mappable selection criteria for “reduced need for water and nitrogen”

• The same curve serves three people: NUE/WUE for the breeder, maturity-corrected Verticillium scores for the pathologist, maturity class for the Tri-State trials

• Cut-tuber imaging with Whitworth for PMTV necrosis, internal defects and sugar ends — plugged into Feldman’s tuber workflow, not a rebuild; separate tolerance from infection with his RT-qPCR

Feldman et al. 2024, Plant Phenome J. 7:e20099 · Navarre, Feldman, Sathuvalli, Charlton, …, Blauer 2026, Am. J. Potato Res. (blackheart) · Yusuf et al. 2025, Plant Genome (UAV multispectral improves genomic prediction) · SGPG FY2025 (PMTV)

Aim 3 · populations · Objectives 2 and 4

Populations that do double duty — starting with the ones Aberdeen already has

• Year-one paper, no field season: one marker set, one reference (DM v6.1 + phased Castle Russet) → an Aberdeen consensus-QTL resource

• New crosses are breeding populations first: Tri-State parents × PCN, PMTV and S. microdontum donors — genotyped at seedling, imaged in the field, no acre spent only on research

Five-year deliverables

• 1 A DArTag-based marker service on every seedling family — PVY, TRV, PCN, PLRV, root-knot — with dosage-aware parent lists and clone-identity QC through Tri-State

• 2 The first maturity-corrected canopy traits for N and water use in potato, with heritabilities, QTL and prediction models; imaged PMTV necrosis scores with the pathologist

• 3 An Aberdeen consensus-QTL resource on DM v6.1, and new PCN / PMTV / wild-donor families built to be mapped from day one

• 4 Pipelines a three-person team runs without a bioinformatician — and PVY-, PCN- and PMTV-resistant russets reaching Tri-State a decision earlier

Year one: DArTag on the next seedling crop · harmonize A08241, A15001, S. microdontum and Eden × Western Russet QTL · fly the early-generation plots and the Aberdeen Tri-State trial · meet the commissions, PVMI and processors · learn the grading line

Summary

Cranberry was the case study. The method is the program. The cycle gets shorter where the decisions do.

References & links

Backup · Zou et al. 2020, Nat. Commun. 11:413 · Clare et al. 2026, Plant Genome 19:e70153

rhAmpSeq and Flex-Seq: two answers to marker transferability

The two recovery figures are not comparable — across a genus is the harder test. And “core” means different things: collinear across species (rhAmpSeq) vs present in every accession (pangenome).

Backup · Endelman et al. 2024 · SGPG FY2024–25 reports · Clare et al. 2026

Potato genotyping today, and where Aberdeen sits

Cranberry: Flex-Seq (17,502 loci) → public DArTag (3,059 loci) → KASP. Potato already has the middle tier — the gap is running it as a service, early.

Backup · Feldman et al. 2024 · ARS Prosser/Wapato FY2025 · Yusuf et al. 2025 · Aono & Chawade 2026

Where potato phenomics stands

• Tuber side is well served: size, shape, skin/flesh color, starch, defects; hollow-heart CNN ~99%; 41 °C / 24 h blackheart induction scored by machine vision (2026)

• UAV multispectral data improves genomic prediction of yield and quality (Yusuf et al. 2025); image-derived environmental kernels match GP under sparse testing (Aono & Chawade 2026)

• Gap: canopy dynamics as genetic traits — heritabilities, QTL, maturity-corrected — and cut-tuber necrosis for PMTV

Backup · "How do you evaluate trait performance?"

Evaluating a trait: three questions, then an economic one

ΔG = i · ρ · σ_A / L — a trait I can measure perfectly that doesn’t change a selection decision isn’t worth measuring

And for a processing russet the economic question is concrete: does it move contract grade, specific gravity, fry color after storage, or sugar ends?

What I am not claiming

• Not a pathologist. PVY, PMTV and TRV biology is Whitworth’s; PCN is Whitworth’s, Kuhl’s and Dandurand’s; my claim is that the numbers the program scores can be continuous, repeatable and mappable

• Not a shorter generation. ΔG = i · ρ · σ_A / L — this moves ρ (accuracy) and the decisions inside L, not tuber-to-tuber biology

• Not yet a polyploid geneticist. Cranberry is diploid; I’ve used polyploid-ready tools on diploid data, not on dosage or tetrasomic problems

• Not a rebuild of Prosser’s imaging. The tuber side exists; the canopy genetics and the Aberdeen populations are the addition

• Not the breeder. The breeder you hire alongside me chooses crosses and selects; my job is the markers, populations and data behind those choices — year one is your populations, your MAS pipeline, your grading line — and the panel’s correction of the phase lengths on the cycle slide

How: open, reproducible, deposited — and paid for

Engineering-grade reproducibility is a research output, not overhead — especially on a team of three.

Backup · spring-greening manuscript, Tables 3–4

LMI rankings: top and bottom 5% per population

Top / bottom 5% (max 10) per population by LMI, random-forest anthocyanin predictions. LMI is population-relative — scores are not comparable across populations.

Predictor importance: parsimony is a feature

GLI dominates — a single green-leaf index carries most of the anthocyanin signal

Predicted A535nm visual progression

Top · Q75 · median · Q25 · bottom of random-forest-predicted A535 — two plot images per quantile, three populations (manuscript Fig. 8)

Prep · get these right

• Role: the Molecular Geneticist position — markers, DNA-guided decisions, populations. Not the breeder.

• Ploidy: cranberry is diploid; Endelman-group tools used on diploid data only. Say it once, plainly.

• Marker timing: FY2025 moved testing later (2nd → 3rd year). Ask why before pitching earlier.

• Feldman: plug in, don’t rebuild — one pipeline across two ARS locations.

• PMTV: all susceptible; all but Castle Russet had some asymptomatic infection. Jonathan’s result.

• PCN: Whitworth leads GLOBAL genetics & breeding — not only UI.

• Pacing: min 16 · 27 · 35 (pivot) · 44 · close. Never cut the vision.

Prep · don’t say

• “Anglin’s former role” (inferred) · Blacker as IPC · “Durrin” by surname — say “the PVMI director”

• The AJPR Outstanding Paper award · any DArTag price · DArTag covers G. pallida (it’s H1)

• Numbers from Dhakal 2026 or Yusuf 2025 · “most PMTV infections asymptomatic”

• Any tetraploid analysis you’ve done · the thesis R² 0.886

• That you’ve worked on quantitative disease resistance — only fruit-rot QTL (not pathogen-specific)

• Do use: Shashi Yellareddygari (WSPC) co-authored the A15001 TRV/PMTV GWAS · aquifer curtailment order updated May 2026

Prep · Q&A in one line each

• Transfer? Same structure; potato has richer infrastructure.

• Dosage? Honest line first, then concepts from the literature.

• Feldman images already? Plug in: canopy genetics, populations, MAS, pathology link.

• MAS or GS? MAS for Ry, TRV chr 9, H1; GS for SG, fry color, G. pallida, PMTV.

• Why earlier than year 3? Ask first; then cost per retained clone + identity QC + GS training from one run.

• Breeder role? Here for the Molecular Geneticist position; my value to the breeder is markers, parent lists, prediction.

• Why phenomics? Markers are only as good as the phenotypes they’re trained on.

• Year one? DArTag on seedlings · harmonize four populations · fly early plots · meet commissions, PVMI, processors.