WHY SOME CROSSES FAIL

This page collects what is actually known. It is short on rules and long on scope, because the subject is one where a confident answer is nearly always someone generalizing from a handful of plants.

Three things are worth knowing before starting. Chromosome counts exist for about a quarter of the family. Four genera on this page have been looked at hard enough to say something, and in four different ways — a breeder's record of what failed, a map of what succeeded, a crossing program of 1,592 hand pollinations across 56 Anthurium species scored by morphological group, and one Zantedeschia cross run to exhaustion with every combination's rate printed. And none of the four is a compatibility table you can look any two species of a genus up in. Outside those four genera the record is not empty: a Cryptocoryne program crossing species from Borneo made 198 attempts, of which 56 set fruit and 17 gave flowering plants.

Two notations turn up. 2n is the count in an ordinary cell of the plant. n is the count in a pollen grain or an egg — half of it. A plant written as 2n = 28 makes pollen carrying 14. When two plants with the same number cross, the halves pair cleanly. When they do not, they may still make a seed, and that seed may still grow — but the plant it becomes often cannot make working pollen of its own.

That is the practical shape of it. A mismatch does not usually stop a cross. It stops the generation after.

How much of the family has been counted

So the first honest answer to “what is my plant's chromosome number” is usually that nobody has looked. The second is that if a number does exist for the species, it may rest on one plant, counted once, decades ago.

The numbers, for the genera people grow

These are Marchant's own counts, made on living plants at Kew between 1970 and 1973 and read here from the printed tables. Where a genus is uniform he counted several species and got the same answer; where it is not, the spread is the interesting part.

Where the seven came from

Between 1970 and 1973 C. J. Marchant worked through the living collection at Kew and published five papers under the title Chromosome variation in Araceae. He set out to reach every genus in the family, and by the last paper he had karyotypes for just over half of them — 44 genera, about a third of the family, by the first paper, with 54 genera simply not available at Kew to be counted at all.

In that final paper he wrote that, with just over half the genera done, x = 7 was the most common basic number, followed by x = 13. That sentence became the figure repeated for the next forty years, though it was not the first time the number had been proposed: a still earlier assessment had already suggested x = 7, or an even lower number.

What was repeated far less often is what he said around it. His scheme of basic-number relationships was published as a diagram whose dotted lines marked “very uncertain derivations”, and he described the relationships as only tentative. He was blunter still about the underlying problem:

There is one more thing worth knowing about the scheme, and it is not about his counts. The review papers that carried this work were organized on a classification of the family that his own field had already rejected. Marchant used the system of Hutchinson — an extension of Hooker's 1883 arrangement, itself built on Schott's 1860 monograph, which sorted the genera into eighteen tribes on floral morphology. Modern workers on Araceae describe that system as “quite unnatural” and it is used by nobody today. The karyotypes were being mapped onto a frame that would not hold.

He was capable of the same scrutiny on himself. The final paper opens its discussion by correcting his own first one: he had concluded in 1970 that Scindapsus pictus, at 2n = ca. 110, had a base number of x = 10, and called that in print “a poor numerical interpretation of the facts”. On a base of x = 7 the same plant is sixteen-ploid.

One further thing his survey settled. An earlier body of work had claimed widespread chromosome variation within species and even within single plants in this family. Marchant's counts gave it no support, and resolved a number of published disagreements as errors rather than biology. That matters for everything below: where two counts for one species disagree, the usual explanation is a mistake, not variation.

Why the seven did not survive

The method had no way of knowing which species were related to which, or how common each number was. It was arithmetic performed on a list. A 1993 reassessment rejected the assumption that the most common number is the ancestral one, favored x = 12, kept x = 13 or x = 14 as live possibilities, found no support for x = 7 or x = 6, and acknowledged that the family's outgroups still could not settle which direction was ancestral. In 2012 the same question was asked again using a model that treats polyploidy, chromosome fusion and chromosome fission as separate events with their own probabilities, fitted to a phylogenetic tree of 113 of the family's 117 genera.

The direction of travel is down

This is the finding that changes how the rest of the page reads. Across the family the model inferred:

Losses outnumber gains by more than eleven to one. The aroid lineage has spent its history shedding chromosomes, not accumulating them — which is exactly the pattern the old highest-common-factor method could not see, because a family that starts high and descends produces a misleadingly small common factor.

Polyploidy is recent, and it is patchy

Twenty-nine polyploidization events were inferred, and they sit mainly at the tips of the tree — in Gymnostachys, Alloschemone, Urospatha, Anubias, Montrichardia, the Cryptocoryneae, Calla, Filarum and Peltandra. Only three are inferred deep in the tree, and two of those are worth naming because they shaped genera people grow: a genome duplication on the branch leading to the Rhaphidophora clade, taking it from n = 15 to n = 30, and a demi-duplication on the branch leading to the Zantedeschia clade, from n = 14 to n = 21.

The Rhaphidophora event is the reason a whole group of familiar climbers — Rhaphidophora, Epipremnum, Scindapsus, Amydrium, Anadendrum and Monstera — all sit at n = 30 while most of the family sits far below it.

And the family itself has moved again

The 2012 re-estimate was fitted to the tree as it stood then. That tree has since been rebuilt on nuclear data rather than chloroplast — target-sequence capture across 128 species in 111 of the family's 143 accepted genera, about 78% of them, by fifteen authors in 2022.

It confirmed the eight subfamilies as monophyletic groups, moved Calla into subfamily Aroideae and absorbed Zamioculcadoideae into it as well. Two things it did not settle are stated plainly in the paper: the position of subfamily Lasioideae remains uncertain, and Montrichardia, Zantedeschia and Anchomanes are only moderately supported.

None of this changes a chromosome count. It changes which plants are being compared when a basic number is calculated — which is exactly the problem this part opened with.

Philodendron: the wall somebody walked into

McColley and Miller divided the genus into three groups by breeding behavior: the arborescent, tree-like plants; the majority of the vines and the true self-headers; and a third set of vine types that had not been crossed with anything, a few of which refused even their own pollen. Plants cross freely within the first two groups. After hundreds of attempts, no successful cross was ever made between them.

They noted in 1965 that the chromosome numbers of their groups were unknown. Marchant counted 2n = 32 in all three of his Kew Philodendron accessions, and a 2026 genomic study of cultivars sorts them into five genetic groups, finds that two of those show no notable hybridization within them while crosses between two others are frequent, and proposes differing chromosome number as the reason, citing earlier counts of 2n = 28 to 40 in Philodendron species.

Alocasia: three clusters, drawn after the fact

An AFLP study fingerprinted 23 cultivars across 17 species and recovered three clusters. The first holds most of the familiar hybrid parents; the second is a smaller group built around Alocasia odora and Alocasia cucullata, with Alocasia portei placed there only through the hybrid ‘Portidora’; the third held a single plant, ‘Hilo Beauty’, so unlike the others that the authors doubted it belonged to the genus at all — and they were right, since in 2009 it was described as Caladium praetermissum.

The finding a hybridizer wants is that almost every documented Alocasia hybrid was developed within a single cluster. The one exception the authors name is Alocasia ‘Novodora’, Alocasia macrorrhizos × Alocasia odora, across the first two. But it is retrospective. It observed hybrids that already existed rather than testing pairs, so it records where the ground has proved firm and says nothing about where it gives way. And the later record shows more. Aroidpedia’s hybrid list gives at least eight further crosses between species of the two clusters, nearly all with Alocasia odora as one parent — ‘Giant Shield’ (Alocasia odora × Alocasia macrorrhizos) and ‘Sanderidora’ (Alocasia odora × Alocasia sanderiana) among them — and a 2025 inheritance study set seed on Alocasia odora with pollen of Alocasia micholitziana ‘Green Velvet’ and Alocasia sanderiana ‘Nobilis’, though never the other way round.

The chromosome numbers sit oddly beside it. Most Alocasia are 2n = 28, but Alocasia odora also occurs as a tetraploid, at 56, and Alocasia longiloba (published as Alocasia lowii) is 70. Two of the genus's most-used parents therefore include plants that are not diploid — and the Alocasia odora plant selfed successfully in a 2025 study was inferred to be a tetraploid from its size and a preliminary flow-cytometry analysis. Whether those two facts are connected is untested and nobody has proposed that they are.

Ploidy claims for trade plants need a measurement behind them. Scott Hyndman wrote on the Aroid-L mailing list in 1998 that he had selected a tetraploid clone of Alocasia reginula 'Black Velvet' from a tissue-culture population, and that tissue culture seemed to produce many of these; he did not say how the ploidy was determined. Alocasia 'Polly', a thicker-leaved sport found among tissue-cultured Alocasia 'Amazonica' liners in the 1980s, was named 'Poly' on the supposition that it was a polyploid, in its finders' account as Steve Lucas relayed it in 2009. Ben Zonneveld measured the DNA content of Alocasia 'Polly' and of the standard plant in 2002 and found it equal, at 10.1 pg, and concluded that Alocasia 'Polly' is not a polyploid. Each is one report about one clone, and only the second reports a measurement.

Anthurium: the counts, and the crosses beside them

Sheffer and Kamemoto counted 63 species, 38 of them for the first time. The most common somatic number is 2n = 30, and the counts range from 20 to 90. Four polyploid series are evident:

Two further complications turn up. B chromosomes — small extra chromosomes outside the normal complement — appear frequently in section Cardiolonchium, in ones and twos and threes. And aneuploidy, an odd chromosome out, shows in at least two species.

The counts have a crossing program beside them. Sheffer and Kamemoto published one in 1976, in a companion paper that reprints these counts as its first table. They sorted 56 species from the University of Hawaii collection into six morphological groups, I–VI, on Englerian characters — ovules per locule, berry color and shape, inflorescence shape, leaf shape and texture — then made 1,592 hand cross- and self-pollinations within and between them. 280 spadices fruited and 181 progenies flowered. The species total is not quite settled: the abstract says 56 and the table lists that many, while the body text twice says 57.

The division between within-group and between-group crosses is where the weight sits. Within a group, 706 pollinations gave 229 fruiting spadices and 172 flowering progenies; between groups, 886 pollinations gave 51 fruiting spadices and 9 flowering progenies, or 1.0%. Set like against like — interspecific crosses only, with the selfs and the within-species crosses taken out — a within-group pollination gave a flowering hybrid 128 times in 633, or 20.2%, against that 1.0% between groups. The authors called that difference significant at the 1% level. One caveat rides with the intergroup figure: the printed total of 9 sits above a column that sums to 8, and only two intergroup pairings produced progenies at all — 1 and 7.

The authors read Group I's low hybrid yield against the spread of chromosome numbers in it, and suggested Group V would have done better still but for the B chromosomes common there. Both are readings offered beside the tables, not tests — the counts and the crosses sit in one paper with no hypothesis joining them.

Two species a grower is likely to own carry their own line in the result. Anthurium andreanum, in the clone 'Uniwai', was put to 51 different interspecific pollinations and produced hybrids only within its own group, with Anthurium concinnatum, Anthurium hoffmannii, Anthurium lindenianum, Anthurium micromystrium, Anthurium nymphaeifolium and Anthurium pichinchae. Anthurium scherzerianum was put to 39 in the same program and produced none.

First, rule out the dull explanation

Everything above is about walls between genomes. The commonest reason a cross fails is duller than any of them: the pollen was dead before it touched the stigma. A wall you cannot cross and pollen that was never viable produce exactly the same result — no fruit — and only one of them is worth writing up.

There is an old and cheap way to tell them apart. Pollen put into a 3% sucrose solution will, if it is alive, swell to a sphere and push out a tube; in Ambrosinia bassii, every grain had produced a long tube within fifteen hours.

In the wild, the wall may lie outside the plant altogether. Peter Boyce wrote on the Aroid-L mailing list in 2009 that aroids in the wild very rarely hybridize, and that the evidence from Asia points less to pollen incompatibility than to niche partitioning: different species use different guilds of pollinators, such as diurnal, crepuscular and nocturnal beetles. It is his reading of the field evidence, given without figures, and hand-pollination bypasses it entirely.

Boyce had singled out Arum as the exception in 2007. In a list of confirmed natural Arum hybrids posted to the same mailing list, he wrote that natural aroid hybrids seem not at all common in nature and that Arum, by contrast, seems profligate. The list includes the hybrid Arum 'Sooi' (Arum cylindraceum × Arum maculatum), first recorded from Hungary and present almost everywhere the two species meet in Europe, and Arum italicum × Arum maculatum, recorded from France. He gave each cross with the pollen parent first, and the pairs are written here in his order. It is a list of occurrences with no counts, so it shows that these pairs do cross in the wild, not how often. The same message also mentions a hybrid in Arisarum and, from Sarawak, hybrids in Alocasia.

One study crossed four Alocasia species and four cultivars in 20 combinations and scored the seedlings for leaf characters. It is the first inheritance data the genus has, and its results are unusually clean.

Every cross matched its expected ratio exactly.

The table gives scored characters in known crosses. A grown-out seedling lot shows what the unscored variation looks like. Ron Weeks, who bought many Alocasia 'Portora' and Alocasia 'Calidora' seedlings and grew them out to choose selections for tissue culture, wrote on the Aroid-L mailing list in 2000 that the plants ranged from lance-leaved miniatures to broad-leaved giants, and that petioles and veins in 'Portora' ran from reddish to pale yellow. Alocasia 'Portora' is itself one selection from an Alocasia portei × Alocasia odora cross made in the early 1980s; with help from Jim Georgusis, Weeks picked a red form and a green form of it, and one form of 'Calidora', for tissue-culture production. His account is a grower's description of what he saw, without counts or the seedlings' pollination records, and not an inheritance ratio.

And beyond Alocasia

One genus, one paper, twenty crosses, leaf characters only — but other genera have leaf genetics of their own. In Caladium cultivars, leaf shape was scored in more than 2,600 seedlings from 38 crosses and behaved as one gene with two co-dominant forms, and main-vein color followed a second, independent gene. A later cross of 'Miss Muffet' × 'Candidum', pooled to 1,267 seedlings, adds a seed-background gene independent of the leaf-shape gene but linked to the genes for main-vein color, spotting and leaf rugosity, in that order along the linkage group. In Dieffenbachia seguine (published as Dieffenbachia maculata), variegation in 'Perfection' and 'Hoffmannii' behaved as a single dominant gene across 781 seedlings from six crosses. Not every variegation has been followed through seedlings: Amorphophallus konjac 'Shattered Glass' arose as a bud sport on a side shoot, no sibling ever showed the trait, and its developer, Michael Marcotrigiano, wrote on the Aroid-L mailing list in 2004 that he read the lethal-white, all-green and variegated corms it gives as segregation of a genetic mosaic rather than a virus. That is his reading of one plant, and the message reports no crosses. A flower character has been followed in Anthurium: scent segregated in the first-generation progeny of all 22 crosses between non-fragrant and fragrant plants, which points to several genes rather than one. How growth habit or thermogenesis is inherited has not been reported for any aroid.

For Philodendron a 2026 study maps the hybridization history of a cultivar collection and works on the mechanism behind leaf color. It is not an inheritance study, and this page does not treat it as one.

Peer-reviewed literature, and mailing-list posts

What is missing

The gaps here are larger than the literature, and they are specific enough that a grower with records could close several of them.

1. A crossing experiment scored species by species. One exists for Anthurium: 1,592 hand pollinations across 56 species, failures and all. But its 344 interspecific cross-combinations are reported by morphological group rather than by pair: roughly a dozen named species pairs carry an outcome of their own, and the other three hundred are group averages. Few other genera have published outcomes to set beside it: a Cryptocoryne program crossing species from Borneo prints 198 attempts, 56 of which set fruit and 17 gave flowering plants, and the Philodendron work prints no per-cross counts at all.

2. Chromosome numbers for three-quarters of the family. Counts exist for about a quarter of Araceae by species. Marchant reached just over half the genera and named the obstacle plainly: 54 genera were not growing at Kew, so they could not be counted. Those genera were closed out over the following decades, the last of them, Pedicellarum, in January 2005, leaving only one genus in the family, Croatiella, still without a published karyotype. The remaining gap is at the species level, not the genus level: for most genera the number in circulation still rests on one or two plants.

3. Whether the breeding groups and the chromosome groups are the same thing. In Philodendron a breeder's barrier and a geneticist's proposed mechanism point the same way, sixty-one years apart. Nobody has tested that they are the same wall.

4. What ploidy does to a cross in this family. Alocasia odora occurs as both a diploid (2n = 28) and a tetraploid (2n = 56), and the plant selfed successfully in a 2025 study was inferred to be a tetraploid; Alocasia lowii is 2n = 70. Both are used as parents. For Alocasia, ploidy has not been tested against crossing success. In Zantedeschia it has: crosses between the genus's two sections were made with diploid and with colchicine-doubled tetraploid parents, and using a tetraploid as one parent did not improve the cross — though the doubled Zantedeschia aethiopica 'Childsiana' plants were later found to be chimeras, part diploid.

5. Inheritance of almost anything beyond the leaf. A flower character has been followed in Anthurium: scent segregated in the first-generation progeny of all 22 crosses between non-fragrant and fragrant plants, pointing to several genes. How growth habit or thermogenesis passes to the next generation has not been reported for any aroid.

6. The failures, almost everywhere. The record of crosses people attempted and lost is the most useful thing missing from this subject. Anthurium has one, which is why the group barrier can be stated as a rate rather than an impression. Zantedeschia has another, and it is built the same way: eleven of its 37 combinations produced no shoot at all, and each is printed with the number of crosses and embryos behind it, so a zero there means somebody tried a stated number of times. A Cryptocoryne program crossing species from Borneo counts its failures too: of 198 attempts, 142 set no fruit. Beyond those, nothing here is built that way, and the rest of the conclusions on this page are weaker for it.