Sources of Contamination

While our selection criteria are designed to identify dust obscured individual stars, there are several classes of contaminating sources as well. QSOs have red mid-IR SEDs compared to stars (e.g., Stern et al.2005), as do star forming galaxies with strong PAH emission at 8.0$\micron$ (e.g. the SED models in Assef et al.2010). Sources in the galaxies such as dusty star clusters and H2 regions can also appear as candidates. We used the SIMBAD and VizieR services to search for previous classifications and near-IR counterparts from the 2MASS point source catalog (Cutri et al.2003). We also noted other significant pieces of information, such as a candidate location coinciding with known radio and X-ray sources. For example, many of the candidates in M33 are associated with known supernova remnants (SNRs). We reject a source as ``non-stellar'' if it is a galaxy or QSO with a measured redshift or if archival images clearly show that it is a galaxy. The rejected sources that meet our selection criteria are described in Section3.1.

We estimate the expected number of extragalactic contamination for each galaxy using the SDWFS survey (Ashby et al.2009) where the nature of the sources, particularly AGNs, is also well understood from the AGES redshift survey (Kochanek et al.2012b). We transform the apparent magnitudes of all sources in a 6deg$^2$ region of SDWFS to luminosity using each target galaxy's distance modulus, determine how many of them would meet our selection criteria, and correct that count for our survey area around each galaxy. Table3 reports the expected surface density of extragalactic contaminants and the number expected given the survey area around each galaxy. We expect a total of $\sim41$ extragalactic sources to pass our selection criteria across the targeted galaxies, as compared to 46 initial candidates. Figure4, which has the same format as Figure2, illustrates this for M81's distance. In the 6deg$^2$ SDWFS area, 449 ($\sim 75$deg$^{-2}$) sources pass our selection criteria, indicating that we should expect $\sim 13$ background sources meeting our selection criteria given our 0.17deg$^2$ survey region around M81, as compared to the 14 initial candidates selected. Indeed, as we discuss in Section4, we can already identify 11 of the 46 initial candidates as extragalactic. Statistically, this means that only $6\pm6$ are likely associated with the galaxies. Also note in Figure4 that none of the contaminating background sources have properties directly comparable to $\eta$Car. The expected numbers of contaminating sources are generally consistent with the observed numbers with the exception of NGC247, which we investigated but appears to surely be a statistical fluke. The angular distribution of the candidates relative to the galaxies is also strongly suggestive of a dominant contribution for background sources.

Table 3: Candidate Statistics

	NGC	M33	NGC	NGC	M81	NGC	NGC
	6822		300	2403		247	7793
Survey Area $A$ (deg$^2$)	0.1	0.73	0.17	0.12	0.17	0.2	0.044
Candidates	0	9	1	5	14	3	14
Expected Background $\Sigma$ (deg$^{-2}$)	0.17	1.67	11	47	75	75	110
Expected Contamination $A\Sigma$	0	1	2	6	13	15	5
Rejected Candidates	0	0	0	0	7	1	3
Remaining Candidates	0	9	1	5	7	2	11

Many of the candidate SEDs show a ``dip'' from 3.6$\micron$ to 4.5$\micron$ before rising again at 5.8$\micron$ (see Figure5). This is a common feature of star cluster SEDs created by strong PAH emission at 3.6$\micron$ (Whelan et al.2011). In total galaxy spectra, this is a weaker effect and the dominant PAH emission feature is at $8\micron$ and comes more from the diffuse ISM rather than individual stars or clusters. The SEDs of $\eta$Car and ObjectX do not show this dip at 4.5$\micron$. We treat the presence of this dip as a qualitative indicator that the source may be a cluster or lie in a cluster. Deep Hubble Space Telescope (HST) images of these regions, where available, can help us distinguish single bright red stars from clusters of fainter stars that may be merged into a single bright source in the lower resolution Spitzer images. Whether these clusters can potentially hide $\eta$Car analogs is discussed in Section2.4.

In Figure7, we present SEDs of four different types of objects that met our selection criteria -- a likely dusty star in NGC2403, a star-cluster in M33, a QSO behind M81, and a galaxy behind NGC7793. Although all four objects met our selection criteria, the detailed shapes of their SEDs are very different from each other. The likely stellar source, N2403-3, has a very steeply rising mid-IR SED that peaks between 8$\micron$ and 24$\micron$. While the compact cluster (M33-8) SED looks quite similar to that of $\eta$Car in the IRAC bands, it continues to rise steeply up to 24$\micron$, and the MIPS 70$\micron$ and 160$\micron$ upper limits show that it peaked between 24$\micron$ and 70$\micron$. The SED of the QSO (M81-4) remains relatively flat from 3.6$\micron$ to 24$\micron$. The MIPS 70$\micron$ and 160$\micron$ upper limits for N7793-2 (a galaxy) are quite stringent, because the source is far from the center of the galaxy, and would rule out an $\eta$Car analog model. For some cases, such as N7793-2, HST images clearly determine the nature of the source (Figure10, bottom panel).